CN102117116B - Moving object recognition method and instruction input method based on moving object recognition - Google Patents
Moving object recognition method and instruction input method based on moving object recognition Download PDFInfo
- Publication number
- CN102117116B CN102117116B CN 200910266219 CN200910266219A CN102117116B CN 102117116 B CN102117116 B CN 102117116B CN 200910266219 CN200910266219 CN 200910266219 CN 200910266219 A CN200910266219 A CN 200910266219A CN 102117116 B CN102117116 B CN 102117116B
- Authority
- CN
- China
- Prior art keywords
- block
- moving object
- correlation
- degree
- time point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
A moving object recognition method is used for recognizing a moving object and marking the position of the moving object. The method comprises the following steps of: capturing continuous dynamic pictures of the moving object, dividing the continuous dynamic pictures into a plurality of blocks; selecting one block and calculating colour characteristic values of the block at the current time point and the next time point; obtaining variability of the block according to the colour characteristic values of the current time point and the next time point; comparing the colour characteristic value of the block at the current time point with the colour characteristic values of the other blocks at the next time point, respectively obtaining a similarity and defining the similarity with maximum value as a local degree of correlation; obtaining a motion intensity value of the block according to the variability of the block and the local degree of correlation; repeating the above steps and obtaining the motion intensity values of all blocks to form a motion intensity picture; and finding out the position of the moving object at the current time point according to the motion intensity picture.
Description
Technical field
The present invention is relevant with the track identification of moving object, particularly about a kind of moving object recognition methods, and the instruction input method of based on motion object identification.
Background technology
Utilize gesture to swing and replace directly contact computer installation, carry out the input of instruction, be a development trend of computer installation.In traditional gesture input system, the user must put on special gloves or fingerstall, utilizes the variation of gloves or fingerstall inducing palm attitude or position, and produces corresponding input instruction.For the inconvenience in gloves or the fingerstall use, the images of gestures recognition technology just is introduced in the gesture input system, utilizes video camera acquisition images of gestures, thereby analyzes the variation of its attitude or position, and produces corresponding input instruction.
Present Gesture Recognition is with the hand contour images through pre-treatment, identifies the variation of palm attitude or position.For example, No. 393629, TaiWan, China patent, I224288 Patent Case utilize different calculation mechanism transformation images, with special characteristic as gesture in image, to find out palm; The TaiWan, China patent illustrated for I298461 number by finding out the static gesture image in the image, again with database in certain gestures image ratio pair.The successful identification of preceding method whether all depend on can be accurately by cutting out the gesture profile in the image or extracting the linear feature of gesture profile.Yet cutting gesture profile and extraction linear feature often are subject to the impact of background, light source and tan alt, simultaneously, the distance of hand and video camera, the attitude of hand itself changes, and also can have influence on the cutting of gesture profile.In order to promote discrimination, often must set up a large amount of default gesture databases for comparison, or increase Error Tolerance.Utilize a large amount of default gesture databases for comparison, can affect recognition speed, must expend relatively many hardware resources, increase the probability that Error Tolerance has then increased the recognition results mistake.
Aforesaid front case is to identify for static hand images, therefore need to carry out the comparison of cutting, linear feature extraction and the database of gesture profile.TaiWan, China patent I274296 patent and U.S. US5594769 patent then are to search dynamic object in continuous dynamic image, with identification dynamic gesture image.But I274296 patent and US5594769 patent are subject to easily, and personnel move in ambient light, the environment, face complexion, camera lens rocks or picture noise impact, and the object beyond the palm is considered as dynamic gesture, and produce erroneous judgement.Therefore, although dynamic gesture identification does not need the accurately cutting of gesture profile, but still must solve the problem that motive objects, noise etc. is mistaken for dynamic gesture.
Summary of the invention
There are respectively the problem that need to set up a large amount of default gesture databases or motive objects, noise etc. is mistaken for dynamic gesture in static gesture identification method and dynamic gesture identification method in the prior art.The present invention proposes a kind of moving object recognition methods, and its degree of accuracy is high and needed operation efficiency is low.
The present invention proposes a kind of moving object recognition methods, in order to identify a moving object, and the position of this moving object of mark, the method comprises the following step: (a) capture the continuous dynamic image of this moving object, and to cut apart this continuous dynamic image be a plurality of blocks; (b) a selected block calculates this block in the color feature value of a current point in time; (c) calculate this block in the color feature value of next time point; (d) in this block, be dependent on the color feature value of this current time point and in the color feature value of next time point, obtain a mobility of this block; (e) with the color feature value of this block in current point in time, compare in the color feature value of next time point with other block one by one, obtain respectively a similarity, and definition to have peaked similarity be the local correlation degree; (f) according to mobility and this local correlation degree of this block, obtain an exercise intensity value of this block; (g) repeating step (b) obtains the exercise intensity value of all those blocks to (f), to form an exercise intensity image; And (h) find out this moving object in the position of current point in time according to this exercise intensity image, wherein this mobility is the absolute value of the color feature value difference of the color feature value of this current time point and this next time point, color feature value divided by this current time point, wherein step (e) also comprises: should be grouped into a plurality of the first blocks and a plurality of the second block by a plurality of blocks, wherein between two adjacent these a plurality of first blocks, at least have one of these a plurality of second blocks, and each this first block is surrounded by those second blocks of part; Define a plurality of the 3rd blocks, respectively the size of the 3rd block and the first block, the second block are identical, and respectively the centre of form of the 3rd block is to be positioned at the sideline mid point of the second block or the end points in sideline; Find out the person that possesses the maximum similarity in these a plurality of first blocks, defining its similarity is the one first block degree of correlation; Search is around this second block of this first block with this first block degree of correlation, finds out the person that possesses the maximum similarity in those second blocks, and to define its similarity be the one second block degree of correlation; Relatively this first block degree of correlation and this second block degree of correlation; If reach this first block degree of correlation greater than this second block degree of correlation, then get this first block degree of correlation and be this local correlation degree.And with this first block with first block degree of correlation as the position of moving object in next time point, if wherein the second block degree of correlation is greater than this first block degree of correlation: search the 3rd block around the second block with this second block degree of correlation, find out the person that possesses the maximum similarity in those the 3rd blocks, and to define its similarity be the 3rd block degree of correlation; Relatively this second block degree of correlation and the 3rd block degree of correlation; If reach this second block degree of correlation greater than the 3rd block degree of correlation, then get this second block degree of correlation and be this local correlation degree, if wherein the 3rd block degree of correlation is greater than this second block degree of correlation, gets the 3rd block degree of correlation and be this local correlation degree.
The present invention also proposes a kind of instruction input method of based on motion object identification, by identifying the change in location of a moving object, produce a movement locus, with input to input instruction that should movement locus, comprise the following step: carry out aforesaid moving object recognition methods, obtain the position of this moving object; Record this position in temporarily providing room; Change in location according to this moving object produces a movement locus; Judge whether this movement locus meets the definition of instruction input; When this movement locus meets the definition of the instruction input of setting, output is to input instruction that should movement locus; Judge that one follows the trail of the state of label; When this tracking label is no, then this tracking label of initialization is yes, and removes this temporarily providing room; When this tracking label is yes, then directly record this position in this temporarily providing room; When this movement locus meets the definition of the instruction input of setting, it is no then will following the trail of label; And work as the definition that this movement locus does not meet the instruction input, then obtain again the position of this moving object.
The present invention integrates the local correlation degree that mobility reaches and the moving object change in location produces that moving object produces, and sets up the exercise intensity image.The exercise intensity image is in order to finding out the moving object in the continuous dynamic image, but and filtering noise, shadow variation etc. may cause the factor of erroneous judgement, and can avoid the color object similar to gesture to be mistaken for gesture.Compared to known technology, the present invention does not need to carry out background removal, so that the tracing step of moving object can be reduced to two independently flow processs, and can indicate rapidly the locus of moving object with relatively low calculation resources.
Description of drawings
Fig. 1 is the process flow diagram of moving object recognition methods of the present invention.
Fig. 2 is for carrying out the system block diagrams of moving object recognition methods of the present invention.
When Fig. 3 and Fig. 4 were current point in time and next time point, dynamic image segmentation was the synoptic diagram of a plurality of blocks continuously.
Fig. 5 is that moving object is when current point in time, in continuous Dynamic Graph the position of image.
Fig. 6 is that moving object is when next time point, in continuous Dynamic Graph the position of image.
Fig. 7 is when current point in time, the distribution plan of exercise intensity image.
Fig. 8 is the process flow diagram of the instruction input method of based on motion object identification.
Fig. 9 is in the one example of the present invention, the process flow diagram of the multimedia playing program of the instruction input method of employing based on motion object identification.
Figure 10 and Figure 11 are the present invention stage by stage in the searching procedure, the first block, the second block, and the synoptic diagram of the 3rd block.
Figure 12 is the present invention stage by stage in the searching procedure, finds out the synoptic diagram of moving object.
[main element label declaration]
10 moving objects
20 data processing equipments
30 image capturing devices
Embodiment
Consult shown in " Fig. 1 " to " Fig. 4 ", a kind of moving object recognition methods that proposes for the embodiment of the invention, in order to identifying a moving object 10, and the position of this moving object 10 of mark, thereby by the change in location of moving object 10 in time shaft, produce the input instruction.Aforesaid moving object 10 can be the operator's of a data processing equipment palm, and this data processing equipment 20 (for example computing machine or notebook computer) is the program of installing, and uses the execution instruction input method.
Consult " Fig. 1 " and reach shown in " Fig. 2 ", moving object recognition methods of the present invention is that this image capturing device 30 can be a video camera by the continuous dynamic image of an image capturing device 30 acquisition moving objects 10, external or in be built in this data processing equipment 20.This image capturing device 30 captures the continuous dynamic image (step 110) of this moving object 10 according to sequential, and should be sent to this data processing equipment 20 by continuous dynamic image.
Image capturing device 30 captures respectively shadow lattice (Frame) in current point in time t-1 and next time point t.Current point in time t-1 and next time point t can be the time point that image capturing device 30 produces two shadow lattice (Frame) continuously.But also a plurality of shadow lattice in interval between current point in time t-1 and next the time point t, that is after current point in time t-1 obtains shadow lattice, obtain again next image behind several shadow lattice of interval, and the time point that produces with these next shadow lattice is as next time point t.
Consult " Fig. 1 ", " Fig. 3 " reaches shown in " Fig. 4 ", then, this data processing equipment 20 should continuous dynamic image segmentation be a plurality of block X (n) (step 120), as shown in the figure, this continuous dynamic image be split into 5x5 totally 25 blocks (X (n) is to X (i=0 of n ± i), ± 1 to ± 12).Aforementioned continuous dynamic image be split into 5x5 totally 25 blocks only be demonstration example of the present invention, be not the cutting quantity that limits continuous dynamic image.
Consult " Fig. 3 " and reach shown in " Fig. 5 ", then data processing equipment 20 selected block X
T-1(n) (step 130) calculates this block X
T-1(n) in the color feature value Vec of current point in time t-1
T-1(n) (step 140).
" Fig. 3 " reaches " Fig. 5 " and gets moving object 10 to be positioned at n block X (n) and to demonstrate, and how to find out moving object 10 in the position of current time t-1 so that moving object recognition methods of the present invention to be described.In fact each block all needs to carry out identical handling procedure, and data processing equipment 20 just can be found out moving object 10.
Consult " Fig. 4 " and reach shown in " Fig. 6 ", data processing equipment 20 calculates this block X
t(n) in the color feature value Vec of next time point t
t(n) (step 150).For convenience of description, in the block of current point in time t-1, be to carry out annotation take (t-1) as subscript, the block of next time point t is to carry out annotation take (t) as subscript.
Moving object 10 except along the two-dimensional directional Linear-moving, in fact also has along approaching or moving away from the direction of image capturing device 30 in continuous dynamic image, causes moving object 10 sizes in the continuous dynamic image to occur changing; In addition, moving object 10 also may be rotated, and its kenel is changed.Particularly, the moving object 10 that captures among the present invention is mainly palm, and the kenel of palm itself just can not fixed.
For the foregoing reasons, when in continuous dynamic image, finding out moving object 10 and judging change in location before and after its motion, must consider moving object 10 in continuous dynamic image distortion and whole continuous dynamic image in affected by light color change.Therefore, the present invention adopts color feature value Vec (n) (color feature value, Vec (n)) feature of each block is described, thereby by the variation of color feature value Vec (n), find out moving object 10 in current point in time t-1 and the possible position of next time point t.
Color feature value Vec (n) can be color variation (color moment), color histogram (color histogram) of each block etc., wherein color variation (color moment) has relatively simple computation process, and therefore the preferred embodiment of aforementioned color feature value Vec (n) is color variation (color moment).
Color feature value Vec (n) has been arranged, and data processing equipment 20 is in this designated blocks X
T-1(n) in, be dependent on the color feature value Vec of this current time point t-1
T-1(n) reach in the color feature value Vec of next time point t
t(n), obtain this block X
T-1(n) a mobility Active (n) (step 160).
Mobility Active (n) is same block (X
T-1(n), X
t(n)) at the color feature value difference (Vec of different time points
T-1(n), Vec
tThat is the color feature value Vec of current point in time t-1 (n)),
T-1(n) change the color feature value Vec of next time point t into
t(n) rate of change.The descriptor format that mobility Active (n) is best is the color feature value Vec of current point in time t-1
T-1(n) with the color feature value Vec of next time point t
t(n) absolute value of difference is divided by the color feature value Vec of current point in time t-1
T-1(n), as follows:
When mobility Active (n) variation is larger, representing this block X (n) may exist the probability of moving object 10 larger, therefore by current point in time t-1 to the time history of next time point t, color feature value Vec (n) has relatively large variation.If mobility Active (n) is minimum, then color feature value Vec (n) may less than variation or rate of change be little, and then representing the image that comprises among this block X (n) is actionless background.
Suppose that moving object 10 is positioned at X when current point in time t-1
T-1(n) in, and when next time point t, move to X in the shadow lattice of next time point t in the shadow lattice of moving object 10 by current point in time t-1
t(n-12) position.Image capturing device 30 was taken a sample with the quite short time interval, generally speaking obtained shadow lattice in about 1/30 second.Judge that according to image capturing device 30 characteristics moving object 10 should be able to move to contiguous block in next time point t.With the color feature value Vec of moving object 10 in current point in time t-1
T-1(n) with next time point t in other block X
t(n+i) color feature value Vec
t(n+i) compare and after obtaining similarity relation.Judge moving object 10 in the possible position of next time point t by similarity, so get final product the noise that flashes in the filtering image.
Though above stated specification is calculated for block X (n), the mobility Active (n) that is actually each block (X (n)～X (n ± i)) needs to obtain.
Therefore, data processing equipment 20 is with this block X
T-1(n) in the color feature value Vec of current point in time t-1
T-1(n), one by one with other block (X
t(n ± i), i=± 1 is to ± 12) are in the color feature value (Vec of next time point t
t(n ± i) i=± 1 to ± 12) compares, obtains respectively a similarity, and definition to have peaked similarity be local correlation degree Corr (n) (step 160).The mathematics kenel of local correlation degree, as follows:
Motion-energy?patch：E(n)＝Active(n)×Corr(n)
Thus, just can judge that when the current point in time t-1, X (n) and two block whichever of X (n-12) are only the block that possesses moving object 10 by exercise intensity value E (n).
As previously mentioned, (X (n) needs to calculate its mobility Active (n), local correlation degree Corr (n) and exercise intensity value E (n) to X (i=0 of n ± i), ± 1 to ± 12) to each block.Therefore data processing equipment 20 repeating steps (step 130) are to step (step 180), obtain all those blocks X (i=0 of n ± i), ± 1 to ± 12) exercise intensity value E (n), use and form an exercise intensity image (Motion-Energy Map), with by finding out moving object 10 in the exercise intensity image.
Consult shown in " Fig. 7 ", the exercise intensity value E (n) of data processing equipment 20 during with current point in time t-1, insert in the matrix of one group of corresponding each block, with fortune degree intensity level E (n) the composition one exercise intensity image (Motion-Energy Map) of all blocks.In " Fig. 7 ", represent high exercise intensity value E (n) with person of light color in the exercise intensity image, the dark person of color represents harmonic motion intensity level E (n), then the exercise intensity image when current time t-1 is shown in " Fig. 7 ".Data processing equipment 20 can be set a threshold value, and one or more block that definition exercise intensity value E (n) surpasses threshold value is this moving object 10 (step 190).Data processing equipment 20 just can be found out moving object 10 in the exercise intensity image when current point in time t-1 thus, and the position of moving object 10.
Reach shown in " Fig. 7 " with reference to " Fig. 3 ", " Fig. 4 " simultaneously, exercise intensity image by current point in time t-1 gained, can be at the block of corresponding X (n), obtain relatively high exercise intensity value E (n), represent when current point in time t-1, moving object 10 is the blocks that are arranged in X (n), and towards other position motion.
Each time point is set as current point in time t-1, implement preceding method one by one after, just can obtain moving object 10 in the position of different time points, and obtain according to this movement locus, thereby carry out instruction input by movement locus.
Consult shown in " Fig. 8 ", based on aforesaid moving object recognition methods, the present invention further proposes a kind of instruction input method, and the change in location by identification moving object 10 produces a movement locus, to carry out input instruction that should movement locus.The data processing equipment 20 of carrying out this instruction input method is to have stored in advance movement locus and the corresponding input instruction of movement locus.
Along with time cumulation, data processing equipment 20 is obtained continuous dynamic image (step 210) according to time shaft, separates the shadow lattice (step 211) of current point in time t-1 and the shadow lattice (step 212) of next time point t.
Then data processing equipment 20 calculates mobility Active (n) and the local correlation degree Corr (n) (step 220) of each block according to aforesaid shadow lattice.
Then data processing equipment 20 is according to aforesaid mobility Active (n) and local correlation degree Corr (n), calculate each block in the exercise intensity value E (n) of current point in time t-1, component movement intensity image according to this, and define one or more block for this moving object 10 according to the exercise intensity image, and obtain the position of this moving object 10.
After the position of moving object 10 had been arranged, data processing equipment 20 was then judged the state of following the trail of label.
If it is no following the trail of label, then data processing equipment 20 will be followed the trail of label and be initialized as and be, and remove its temporarily providing room, with setting in motion trajectory track flow process (step 251); The position that then continues record moving object 10 if yes produces movement locus in this temporarily providing room with the variation by the position.
After the flow process of the judgement movement locus of step step 251 began, data processing equipment 20 was recorded in (when the current point in time t-1) position of moving object 10 in the temporarily providing room (step 260) first.According to the movement locus that the change in location of moving object 10 produces, data processing equipment 20 can judge just whether movement locus meets the definition (step 270) of instruction input.
Because data processing equipment 20 just will be followed the trail of label and be initialized as and be this moment, and removing temporarily providing room, therefore only have the data that store a position in the temporarily providing room, can't produce track to meet the definition of arbitrary instruction input, therefore loop back and be grouped into step step 210, make data processing equipment 20 obtain again moving object 10 in the position of follow-up time point.
On the contrary, if when step step 240, it has been yes following the trail of label, then represent and recorded previous obtained moving object 10 positions in the temporarily providing room, at this moment, data processing equipment 20 directly enters tracking state (step 252), and directly the position of moving object 10 is recorded in (step 260) in the temporarily providing room, and compare with other position that has been stored in the temporarily providing room, produce the movement locus of moving object 10.
At this moment, meet the definition (step 270) of predefined instruction input if data memory device 20 is judged the movement locus of moving objects 10, then will to follow the trail of label be no to data memory device 20, end tracking state (step 280).Simultaneously, data memory device 20 output is to input instruction (step 290) that should movement locus.If data memory device 20 judges that aforesaid movement locus does not still meet the definition of instruction input, therefore loop back and be grouped into step step 210, make data processing equipment obtain again moving object 10 in the position of follow-up time point, the movement locus of record object 10.
Consult shown in " Fig. 9 ", the example of below lifting a multimedia playing program further specifies this instruction input method.
Multimedia playing program is installed in this data memory device 20, simultaneously, one inputting interface program also is installed in this data memory device 20, and in order to carry out the instruction input mode of based on motion object identification method, this inputting interface program can be integrated in this multimedia playing program.
When the user starts this multimedia playing apparatus (step 310) in data memory device 20, be to start simultaneously this inputting interface program (step 410).This inputting interface program start image capturing device 30, the continuous dynamic image of obtaining to obtain this image capturing device 30.
All be judged as moving object 10 for fear of all by the motive objects that image capturing device 30 is captured, cause the gesture identification operating function of multimedia playing program by false triggering, can add one in the inputting interface program and judge circulation, with the starting point of a specific dynamic behavior as the gesture tracking, namely after the object of specific dynamic behavior occurs, the inputting interface program just begins to carry out the step step 210 shown in " Fig. 8 ", tracking is also recorded the moving object 10 that this image capturing device 30 captures, and judge whether its movement locus meets default gesture.
Consult shown in " Fig. 9 ", in this example, the specific dynamic behavior is to be set as gesture back and forth to brandish, and continues a special time (for example two seconds).That is to say, when the user will enable gesture operation function, be in continuing to brandish more than 2 seconds before the image capturing device 30 with its palm.In the exercise intensity image, aforementioned palm (moving object 10) is by a small margin back and forth brandished, can fixedly produce relatively high exercise intensity value E (n) distribution in block or the fixed range block, in the image that data memory device 20 judgement image capturing devices 30 capture, when continuous gesture occurring and brandishing state more than two seconds (step 320), the status indication of then continuous gesture being brandished more than two seconds is the starting point (step 420) of the movement locus of tracing movement object 10, at this moment the inputting interface program is switched to the step step 210 shown in " Fig. 8 ".
Then, the user just can be in image capturing device 30 front swing certain gestures (step 330), data memory device 20 can sequentially record the position of moving object 10 (that is palm of user), and analyze the movement locus of moving object 10 and record movement locus (step 430) according to change in location.
Just can static gesture (step 340) after the user finishes gesture, make the inputting interface program can't capture again moving object 10.At this moment, whether the movement locus followed the trail of of inputting interface procedure identification meets predefined gesture (step 440).
At last, data processing equipment 20 is finished control step (step 350), and sends corresponding input instruction (step 450), if fully without meeting predefined gesture, then points out None-identified or without definition.
At last, program revert to original loop, waits for that continuous gesture (step 320) occurs.After the user closes multimedia playing program (step 360), also close simultaneously inputting interface program (step 460), catch the normal operation that gesture dynamically affects data memory device 20 to avoid the inputting interface program to continue.
The present invention is a plurality of blocks with continuous dynamic image segmentation, with the position of tracing movement object 10.But this moving object 10 may occupy the part of a plurality of adjacent block simultaneously, so that produce error when judging local correlations, between the position that causes moving object 10 essence positions and system to be judged relatively large drop is arranged, the energy that affects the exercise intensity image is estimated.If diminish yet will cut block, the quantity that needs to compare increases, and causes operand to increase, and must have the relatively high hardware of operation efficiency to begin to carry out the analysis of continuous dynamic image.
In order to address the above problem, the present invention further proposes stage by stage searching procedure, in order to the local correlations of calculation block, to increase again search area when not increasing operand.
Consult shown in " Figure 10 ", carry out before local correlation degree Corr (n) searches, data processing equipment 20 is grouped into those blocks first a plurality of the first blocks 1 and a plurality of the second block 2.Aforementioned the first block 1 and the second block 2 are to be two-dimensional matrix to arrange, and each first block 1 and the second block 2 are the square type.Between 2 first blocks 1 of same dimension, have at least one second block 2, and each first block 1 is surrounded by eight the second blocks.
Consult shown in " Figure 11 ", then, a plurality of the 3rd blocks 3 of data processing equipment 20 definition, respectively the size of the 3rd block 3 and the first block 1, the second block are identical.Again, the centre of form of each the 3rd block 3 is to be positioned at the sideline mid point of the first block 1 or the second block or the end points in sideline.
Consult shown in " Figure 12 ", when checking local degree of correlation Corr (n), data processing equipment 20 is prior to searching, find out the person n1 that possesses the maximum similarity in the first block 1 in those first blocks 1, defining its similarity is the first block degree of correlation Corr (n1):
Corr(n1)＝max{Sim<Vec
t-1(n)，Vec
t(1)>}
Then, the second block 2 that data processing equipment 20 is searched around the first block n1 with first block degree of correlation Corr (n1), find out the person n2 that possesses the maximum similarity in those second blocks 2, and to define its similarity be the second block degree of correlation Corr (n2):
Corr(n2)＝max{Sim<Vec
t-1(n)，Vec
t(2)>}
Otherwise, if the second block degree of correlation Corr (n2) is greater than this first block degree of correlation Corr (n1), then data processing equipment 20 continues to search the 3rd block 3 around the second block n2 with second block degree of correlation Corr (n2), find out the person n3 that possesses the maximum similarity in those the 3rd blocks 3, and to define its similarity be the 3rd block degree of correlation Corr (n3):
Corr(n3)＝max{Sim<Vec
t-1(n)，Vec
t(3)>}
Otherwise, if the 3rd block degree of correlation Corr (n3) is greater than this second block degree of correlation Corr (n2), then to get the 3rd block degree of correlation Corr (n3) be this local correlation degree Corr (n) to data processing equipment 20, and with this 3rd block n3 with the 3rd block degree of correlation Corr (n3) as the position of moving object 10 in next time point t.The 3rd block n3 with the 3rd block degree of correlation Corr (n3), to be overlapped in the first adjacent block 1 and the second block 2, therefore, during with the position of this 3rd block n3 Describing Motion object 10, this moving object 10 be can present and the first adjacent block 1 and the second block 2 occupied fifty-fifty, therefore the physical location of moving object 10 still can be approached in the position of the 3rd block n3 institute Describing Motion object 10.
The present invention integrates the local correlation degree Corr (n) that mobility Active (n) reaches and moving object 10 change in location produce that moving object 10 produces, and uses the moving object 10 of finding out in the continuous dynamic image, and its position of mark.Compared to known technology, the present invention does not need to carry out background removal, so that the tracing step of moving object 10 can be reduced to two independently flow processs, and can indicate rapidly the locus of moving object 10 with relatively low calculation resources.
The present invention considers in the actual conditions that gesture controls in addition, exist the color change that gesture is out of shape and background light causes in continuous dynamic image, so the present invention becomes a plurality of blocks to be described continuous dynamic image segmentation, use the variation in the tolerance actual conditions, also more quick in the computing.
Claims (6)
1. moving object recognition methods, in order to identifying a moving object, and the position of this moving object of mark, the method comprises the following step:
(a) capture the continuous dynamic image of this moving object, and to cut apart this continuous dynamic image be a plurality of blocks;
(b) a selected block calculates this block in the color feature value of a current point in time;
(c) calculate this block in the color feature value of next time point;
(d) in this block, be dependent on the color feature value of this current time point and in the color feature value of next time point, obtain a mobility of this block;
(e) with the color feature value of this block in current point in time, compare in the color feature value of next time point with other block one by one, obtain respectively a similarity, and definition to have peaked similarity be the local correlation degree;
(f) according to mobility and this local correlation degree of this block, obtain an exercise intensity value of this block;
(g) repeating step (b) obtains the exercise intensity value of all those blocks to (f), to form an exercise intensity image; And
(h) find out this moving object in the position of current point in time according to this exercise intensity image,
Wherein this mobility is the absolute value of the color feature value difference of the color feature value of this current time point and this next time point, divided by the color feature value of this current time point,
Wherein step (e) also comprises:
Should be grouped into a plurality of the first blocks and a plurality of the second block by a plurality of blocks, wherein between two adjacent these a plurality of first blocks, have at least one of this a plurality of second blocks, and each this first block be surrounded by those second blocks partly;
Define a plurality of the 3rd blocks, respectively the size of the 3rd block and the first block, the second block are identical, and respectively the centre of form of the 3rd block is to be positioned at the sideline mid point of the second block or the end points in sideline;
Find out the person that possesses the maximum similarity in these a plurality of first blocks, defining its similarity is the one first block degree of correlation;
Search is around this second block of this first block with this first block degree of correlation, finds out the person that possesses the maximum similarity in those second blocks, and to define its similarity be the one second block degree of correlation;
Relatively this first block degree of correlation and this second block degree of correlation; And
If this first block degree of correlation, is then got this first block degree of correlation greater than this second block degree of correlation for this local correlation degree, and with this first block with first block degree of correlation as the position of moving object in next time point,
If wherein the second block degree of correlation is greater than this first block degree of correlation:
Search is around the 3rd block of the second block with this second block degree of correlation, finds out the person that possesses the maximum similarity in those the 3rd blocks, and to define its similarity be the 3rd block degree of correlation;
Relatively this second block degree of correlation and the 3rd block degree of correlation; And
If, then getting this second block degree of correlation greater than the 3rd block degree of correlation, this second block degree of correlation is this local correlation degree,
If wherein the 3rd block degree of correlation is greater than this second block degree of correlation, gets the 3rd block degree of correlation and be this local correlation degree.
2. moving object recognition methods according to claim 1, wherein this current time point and this next time point produce the time point of two shadow lattice continuously.
3. moving object recognition methods according to claim 1, wherein this current time point and this a plurality of shadow lattice in next time point interval.
4. moving object recognition methods according to claim 1, wherein this color feature value is color variation or the color histogram of each block.
5. moving object recognition methods according to claim 1, the step of wherein finding out this moving object according to this exercise intensity image comprise one or more block that definition exercise intensity value surpasses a threshold value and are this moving object.
6. the instruction input method of a based on motion object identification by identifying the change in location of a moving object, produces a movement locus, to input instruction that should movement locus, comprises the following step with input:
Carry out moving object recognition methods as claimed in claim 1, obtain the position of this moving object;
Record this position in temporarily providing room;
Change in location according to this moving object produces a movement locus;
Judge whether this movement locus meets the definition of instruction input;
When this movement locus meets the definition of the instruction input of setting, output is to input instruction that should movement locus;
Judge that one follows the trail of the state of label;
When this tracking label is no, then this tracking label of initialization is yes, and removes this temporarily providing room;
When this tracking label is yes, then directly record this position in this temporarily providing room;
When this movement locus meets the definition of the instruction input of setting, it is no then will following the trail of label; And
When this movement locus does not meet the definition that instruction is inputted, then obtain again the position of this moving object.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910266219 CN102117116B (en) | 2009-12-30 | 2009-12-30 | Moving object recognition method and instruction input method based on moving object recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910266219 CN102117116B (en) | 2009-12-30 | 2009-12-30 | Moving object recognition method and instruction input method based on moving object recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102117116A CN102117116A (en) | 2011-07-06 |
| CN102117116B true CN102117116B (en) | 2013-02-13 |
Family
ID=44215916
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910266219 Active CN102117116B (en) | 2009-12-30 | 2009-12-30 | Moving object recognition method and instruction input method based on moving object recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102117116B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103167290A (en) * | 2013-04-01 | 2013-06-19 | 深圳市云宙多媒体技术有限公司 | Method and device for quantizing video coding movement intensity |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102221891A (en) * | 2011-07-13 | 2011-10-19 | 广州视源电子科技有限公司 | Method and system for realizing optical image gesture recognition |
| CN103576845B (en) * | 2012-08-06 | 2018-02-02 | 原相科技股份有限公司 | Interactive apparatus of the ambient light detection device with method and using this ambient light detection device |
| CN106228109A (en) * | 2016-07-08 | 2016-12-14 | 天津大学 | A kind of action identification method based on skeleton motion track |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6363160B1 (en) * | 1999-01-22 | 2002-03-26 | Intel Corporation | Interface using pattern recognition and tracking |
| CN101312524A (en) * | 2007-05-23 | 2008-11-26 | 财团法人工业技术研究院 | Moving object detecting apparatus and method using light track analysis |
| CN101424972A (en) * | 2007-11-02 | 2009-05-06 | 财团法人工业技术研究院 | Method for calculating position of unstructured object in continuous image |
-
2009
- 2009-12-30 CN CN 200910266219 patent/CN102117116B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6363160B1 (en) * | 1999-01-22 | 2002-03-26 | Intel Corporation | Interface using pattern recognition and tracking |
| CN101312524A (en) * | 2007-05-23 | 2008-11-26 | 财团法人工业技术研究院 | Moving object detecting apparatus and method using light track analysis |
| CN101424972A (en) * | 2007-11-02 | 2009-05-06 | 财团法人工业技术研究院 | Method for calculating position of unstructured object in continuous image |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103167290A (en) * | 2013-04-01 | 2013-06-19 | 深圳市云宙多媒体技术有限公司 | Method and device for quantizing video coding movement intensity |
| CN103167290B (en) * | 2013-04-01 | 2016-05-11 | 深圳市云宙多媒体技术有限公司 | A kind of Video coding exercise intensity quantization method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102117116A (en) | 2011-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11379696B2 (en) | Pedestrian re-identification method, computer device and readable medium | |
| Zhang et al. | Learning regression and verification networks for long-term visual tracking | |
| TWI476632B (en) | Method for moving object detection and application to hand gesture control system | |
| JP5845002B2 (en) | Image processing apparatus and method, and program | |
| CN103415825B (en) | System and method for gesture identification | |
| Yun et al. | An automatic hand gesture recognition system based on Viola-Jones method and SVMs | |
| CN108960163B (en) | Gesture recognition method, device, equipment and storage medium | |
| CN105353634A (en) | Household appliance and method for controlling operation by gesture recognition | |
| US9582711B2 (en) | Robot cleaner, apparatus and method for recognizing gesture | |
| JP2014137818A (en) | Method and device for identifying opening and closing operation of palm, and man-machine interaction method and facility | |
| CN111680594A (en) | Augmented reality interaction method based on gesture recognition | |
| CN110751232A (en) | Chinese complex scene text detection and identification method | |
| EP3090382A1 (en) | Real-time 3d gesture recognition and tracking system for mobile devices | |
| Zhu et al. | Deep residual text detection network for scene text | |
| CN113449610A (en) | Gesture recognition method and system based on knowledge distillation and attention mechanism | |
| CN102117116B (en) | Moving object recognition method and instruction input method based on moving object recognition | |
| CN110533661A (en) | Adaptive real-time closed-loop detection method based on characteristics of image cascade | |
| Avola et al. | A shape comparison reinforcement method based on feature extractors and f1-score | |
| US20220300774A1 (en) | Methods, apparatuses, devices and storage media for detecting correlated objects involved in image | |
| CN112686122B (en) | Human body and shadow detection method and device, electronic equipment and storage medium | |
| CN107797748A (en) | Dummy keyboard input method and device and robot | |
| Mu et al. | Salient object detection in low contrast images via global convolution and boundary refinement | |
| CN111651038A (en) | Gesture recognition control method based on ToF and control system thereof | |
| Bao et al. | A new approach to hand tracking and gesture recognition by a new feature type and HMM | |
| Zhu et al. | Scene text relocation with guidance |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |