CN109859239B - A kind of method and apparatus of target tracking - Google Patents
A kind of method and apparatus of target tracking Download PDFInfo
- Publication number
- CN109859239B CN109859239B CN201910365924.4A CN201910365924A CN109859239B CN 109859239 B CN109859239 B CN 109859239B CN 201910365924 A CN201910365924 A CN 201910365924A CN 109859239 B CN109859239 B CN 109859239B
- Authority
- CN
- China
- Prior art keywords
- frame
- prediction block
- weight
- motion state
- current image
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000001514 detection method Methods 0.000 claims abstract description 144
- 239000013598 vector Substances 0.000 claims abstract description 75
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 238000003062 neural network model Methods 0.000 claims abstract description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 44
- 238000004590 computer program Methods 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 10
- 238000007689 inspection Methods 0.000 claims description 10
- 230000035772 mutation Effects 0.000 abstract description 3
- 238000012937 correction Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 238000012545 processing Methods 0.000 description 6
- 238000005352 clarification Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 239000000700 radioactive tracer Substances 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method and apparatus of target tracking, for carrying out trajectory track for the commodity of motion state mutation, realizes and take the accurate tracking of end article to user.This method comprises: obtaining current image frame and the corresponding prediction block of target and detection block in the current image frame;By preparatory trained neural network model, the second feature vector of target in the first eigenvector of target and detection frame region in the prediction frame region is extracted;It will test frame to be matched with the band of position of the prediction block in the current image frame, and the first eigenvector matched with second feature vector, current motion state is determined according to matching result;Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block.
Description
Technical field
The present invention relates to field of artificial intelligence more particularly to a kind of method and apparatus of target tracking.
Background technique
With the rapid development of artificial intelligence, target detection tracer technique is increasingly taken seriously, and is applied especially to mesh
Mark tracking etc. of vehicle at the uniform velocity or variable motion scene.When carrying out target detection tracking, traditional method is to utilize Kalman
Filtering algorithm predicts the motion conditions of target object, according to the shape of the status predication present frame of target object former frame
State, and be corrected using Current observation state, to guarantee output state smooth change, it is effective against observation error.
When carrying out target detection tracking to moving target, the first frame image of target is provided by simple matching method
Previous frame image first just is utilized using Kalman filter since the state of third frame image with the state of the second frame image
The state of current frame image is predicted, then the testing result of current frame image is used to be corrected as observation, obtained correction
As a result target is taken as in the time of day of current frame image.
But detection tracking is carried out to target object using Kalman filtering algorithm, it is suitable at the uniform velocity or variable motion field
Scape, under the intelligent public safety such as intelligent sales counter or unmanned supermarket, since the action trail for commodity of taking when user's shopping is deposited
It is being mutated, the tracking of object action trail cannot be being carried out according to uniform variable motion model, therefore cannot be using above-mentioned algorithm to jump
The object that frame, motion state are mutated carries out trajectory track.
Summary of the invention
The present invention provides a kind of method and apparatus of target tracking, are applied to the intelligent intelligence such as sales counter or unmanned supermarket
Under public safety, trajectory track is carried out for the commodity of frame-skipping, motion state mutation, realizes and takes the standard of end article to user
Really tracking.
In a first aspect, the present invention provides a kind of method of target tracking, this method comprises:
Obtain current image frame and the corresponding prediction block of target and detection block in the current image frame;
By preparatory trained neural network model, extract in the prediction frame region first eigenvector of target and
Detect the second feature vector of target in frame region;
It will test frame to be matched with the band of position of the prediction block in the current image frame, and by the fisrt feature
Vector is matched with second feature vector, determines current motion state according to matching result;
Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block.
As an alternative embodiment, utilizing the prediction using correcting mode corresponding with current motion state
Frame is corrected the detection block, comprising:
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state
With the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, further include:
Determine current motion state be uniform variable motion state when, determine corresponding first weight of uniform variable motion state and
Second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first weight
Account for again than difference no more than preset value;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, determining current motion state according to matching result, comprising:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the fisrt feature to
When amount is with secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, current motion state is determined
For uniform variable motion state.
As an alternative embodiment, will test the band of position of frame and prediction block in the current image frame into
Row matching, and the first eigenvector is matched with second feature vector, comprising:
Using Hungary matching algorithm, the band of position progress of frame and prediction block in the current image frame will test
Match;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
As an alternative embodiment, will test frame and prediction block described current using Hungary matching algorithm
The band of position in picture frame is matched, comprising:
Calculate the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, really
Determine detection block to match with the band of position of the prediction block in the current image frame.
As an alternative embodiment, using Feature Correspondence Algorithm, by first eigenvector and second feature vector
It is matched, comprising:
The Feature Correspondence Algorithm is European between the first eigenvector and the second feature vector for calculating
Distance, when determining that the Euclidean distance is within the scope of pre-determined distance, determining and the first eigenvector and second feature vector
Corresponding prediction block and detection block matching.
As an alternative embodiment, obtaining the corresponding prediction block of target image and the inspection in the current image frame
Survey frame, comprising:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
As an alternative embodiment, obtaining the corresponding prediction block of target image and the inspection in the current image frame
Survey frame, comprising:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection in the current image frame are obtained
Frame.
Second aspect, the present invention provide a kind of equipment of target tracking, which includes: processor and memory,
In, the memory is stored with program code, when said program code is executed by the processor, so that the processor is used
In execution following steps:
Obtain current image frame and the corresponding prediction block of target and detection block in the current image frame;
By preparatory trained neural network model, extract in the prediction frame region first eigenvector of target and
Detect the second feature vector of target in frame region;
It will test frame to be matched with the band of position of the prediction block in the current image frame, and by the fisrt feature
Vector is matched with second feature vector, determines current motion state according to matching result;
Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block.
As an alternative embodiment, the processor is specifically used for:
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state
With the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the processing implement body is also used to:
Determine current motion state be uniform variable motion state when, determine corresponding first weight of uniform variable motion state and
Second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first weight
Account for again than difference no more than preset value;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the processor is specifically used for:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the fisrt feature to
When amount is with secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, current motion state is determined
For uniform variable motion state.
As an alternative embodiment, the processor is specifically used for:
Using Hungary matching algorithm, the band of position progress of frame and prediction block in the current image frame will test
Match;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
As an alternative embodiment, the processor is specifically used for:
Calculate the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, really
Determine detection block to match with the band of position of the prediction block in the current image frame.
As an alternative embodiment, the processor is specifically used for:
The Feature Correspondence Algorithm is European between the first eigenvector and the second feature vector for calculating
Distance, when determining that the Euclidean distance is within the scope of pre-determined distance, determining and the first eigenvector and second feature vector
Corresponding prediction block and detection block matching.
As an alternative embodiment, the processor is specifically used for:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
As an alternative embodiment, the processor is specifically used for:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection in the current image frame are obtained
Frame.
The third aspect, the present invention provide the equipment of another target tracking, the equipment include: obtain module, extraction module,
Matching module and correction module, in which:
Module is obtained, for obtaining the corresponding prediction block of target and inspection in current image frame and the current image frame
Survey frame;
Extraction module, for extracting target in the prediction frame region by preparatory trained neural network model
The second feature vector of target in first eigenvector and detection frame region;
Matching module is matched for will test frame with the band of position of the prediction block in the current image frame, and
The first eigenvector is matched with second feature vector, current motion state is determined according to matching result;
Correction module, for using correcting mode corresponding with current motion state, using the prediction block to the inspection
Frame is surveyed to be corrected.
As an alternative embodiment, the correction module is specifically used for:
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state
With the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the equipment further includes the second correction module, it is used for:
Determine current motion state be uniform variable motion state when, determine corresponding first weight of uniform variable motion state and
Second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first weight
Account for again than difference no more than preset value;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the matching module is specifically used for:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the fisrt feature to
When amount is with secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, current motion state is determined
For uniform variable motion state.
As an alternative embodiment, the matching module is specifically used for:
Using Hungary matching algorithm, the band of position progress of frame and prediction block in the current image frame will test
Match;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
As an alternative embodiment, the matching module is specifically used for:
Calculate the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, really
Determine detection block to match with the band of position of the prediction block in the current image frame.
As an alternative embodiment, the matching module is specifically used for:
The Feature Correspondence Algorithm is European between the first eigenvector and the second feature vector for calculating
Distance, when determining that the Euclidean distance is within the scope of pre-determined distance, determining and the first eigenvector and second feature vector
Corresponding prediction block and detection block matching.
As an alternative embodiment, the acquisition module is specifically used for:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
As an alternative embodiment, the acquisition module is specifically used for:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection in the current image frame are obtained
Frame.
Fourth aspect, the present invention provide a kind of computer storage medium, are stored thereon with computer program, which is located
The step of reason device realizes above-mentioned first aspect the method when executing.
A kind of method and apparatus of target tracking provided by the invention, has the advantages that
Since the prediction block and detection block of generation when the embodiment of the present invention not only carries out detecting and tracking to target match,
Clarification of objective vector in the prediction block and detection block can also be extracted, to clarification of objective in the prediction block and detection block
Vector is matched, and using matched twice as a result, it is possible to the current motion state of target is determined more accurately, can be applicable in
In multi-motion mode, target following effectively is carried out to different motion modes.
In particular, method provided in an embodiment of the present invention, which is applied to intelligent sales counter, can generate effective tracking effect, it is right
The motion state of mutation when user takes article by intelligent sales counter, is effectively played in real time to the article taken
Tracking.
Detailed description of the invention
Fig. 1 is a kind of method flow diagram of target tracking provided in an embodiment of the present invention;
Fig. 2 is a kind of specific method flow diagram of target tracking provided in an embodiment of the present invention;
Fig. 3 is a kind of equipment schematic diagram of target tracking provided in an embodiment of the present invention;
Fig. 4 is the equipment schematic diagram of another target tracking provided in an embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
It is described in detail to one step, it is clear that described embodiments are only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
All other embodiment, shall fall within the protection scope of the present invention.
Embodiment one
Based on existing target tracking technology, target usually is carried out both for the uniform variable motion target such as vehicle tracking and is chased after
Track, if but existing tracer technique be applied in non-uniform variable motion scene, possibly can not be normally to the mesh in the scene
Mark is tracked, and is easy with losing target.A kind of method of target tracking provided in an embodiment of the present invention, is able to solve above-mentioned technology
Problem.
As shown in Figure 1, this method specifically includes the following steps:
Step 10: obtaining current image frame and the corresponding prediction block of target and detection block in the current image frame.
The present invention can be, but not limited to obtain current image frame by monocular cam.
The present invention can obtain current image frame every preset time, can also obtain current image frame in real time, the present invention
The time and time interval that obtain picture frame are not limited excessively.
Specifically, the embodiment of the present invention obtains the side of the corresponding prediction block of target and detection block in the current image frame
Formula is as follows:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
The quantity of target determines in how much described images for being detected according to detection block of the prediction block, if the inspection
It surveys frame and detects a target, then the prediction block generated is one, if the detection block detects multiple targets, give birth to
At prediction block be it is multiple.
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
Optionally, the embodiment of the present invention can also be based on Kalman filtering algorithm, obtain the mesh in the current image frame
The corresponding prediction block of logo image and detection block.
The algorithm for how obtaining prediction block and detection block is not limited excessively in the embodiment of the present invention, it can be by a variety of
Algorithm of target detection and target tracking algorism obtain.
The quantity for the target image in current image frame that the detection block detects, and it is actual in the target image
The performance of destination number and algorithm of target detection (model) is related.
Step 11: by preparatory trained neural network model, extract target in the prediction frame region first is special
It levies vector and detects the second feature vector of target in frame region.
The embodiment of the present invention does not limit the structure of the trained neural network model in advance excessively, wherein base
In the network structure of neural net base, a variety of neural network models can extract clarification of objective vector in image-region.
Specifically, preparatory trained neural network model described in the embodiment of the present invention includes but is not limited to lower die
Type:
The more frames of single step detect (Single-Shot Detector, SSD) model;AM-Softmax(Additive Margin
Softmax) Feature Selection Model.
Above-mentioned first eigenvector, second feature vector can indicate clarification of objective in corresponding region, such as take
It takes in the scene of article, the target in corresponding detection frame region is article, but the target in corresponding prediction frame region is the article
When the background of place scene, the second spy of the first eigenvector of target and target in detection frame region in the prediction frame region
It is inevitable different to levy vector.This method provided by the invention can confirm current goal from side according to the feature vector in region
Whether movement is mutated.
Step 12: will test frame and matched with the band of position of the prediction block in the current image frame, and will be described
First eigenvector is matched with second feature vector, determines current motion state according to matching result.
Specifically, determining that current motion state has following two according to matched result:
1) determine that the band of position of detection block and prediction block in the current image frame mismatches, and the fisrt feature
When vector and secondary features vector matching, determine that current motion state is non-uniform variable motion state;
Wherein, the non-uniform variable motion state includes but is not limited to the motion state being mutated.
Band of position progress in the embodiment of the present invention to the detection block and prediction block in the current image frame
Match, a variety of matching algorithms can be used, is mainly used for measuring prediction block and detection block in the overlapping region of the current frame image
Number.Different matching algorithms would generally be determined according to corresponding matching degree carries out whether the matched band of position matches,
Generally, it is considered that the detection block and prediction block are in the current image frame when calculated matching degree is less than setting value
The band of position mismatches.
Likewise, the first eigenvector is matched with second feature vector, can using a variety of calculating multidimensional to
Matching algorithm between amount is mainly used for measuring whether feature vector identical or similarity degree of feature vector.Usually recognize
For when calculated matching degree is less than setting value, the first eigenvector and second feature vector are mismatched.When the inspection
The band of position in the current image frame of frame and prediction block is surveyed to mismatch, and the first eigenvector and second feature to
When amount mismatches, current motion state can not be judged.
2) when determining that detection block is matched with the band of position of the prediction block in the current image frame, current kinetic shape is determined
State is uniform variable motion state.
Wherein, the uniform variable motion state includes but is not limited to uniform motion, uniformly accelerated motion and uniformly retarded motion.
3) determine that the band of position of detection block and prediction block in the current image frame mismatches, and the fisrt feature
When vector and second feature vector mismatch, the prediction block is initialized or abandoned.
Specifically, the prediction block is initialized, according to the centre coordinate of the detection block, width and height initialization one
A new prediction block;
According to situation as follows, the prediction block is abandoned:
The prediction block is in uncertain (tentative) state, then abandons the prediction block;
Alternatively, the prediction block is in (confirmed) state of determination, and the prediction block is in preceding setting number of images frame
In all without and detection block match, then abandon the prediction block.Before setting number of images frame as current image frame before described
The picture frame for setting quantity, can set quantity according to demand, by the prediction block and detection block in the picture frame of preceding setting quantity
It is matched, current predictive frame is abandoned if all mismatching.
As an alternative embodiment, will test the band of position of frame and prediction block in the current image frame into
Row matching, and the first eigenvector is matched with second feature vector, include but is not limited to any of the following mode:
Mode one: Hungary matching algorithm is utilized, will test the position area of frame and prediction block in the current image frame
Domain is matched;
The Hungary matching algorithm provided in this implementation is a kind of traditional according to figure matching algorithm, it is therefore an objective to from several
It is a want to obtain in matched target with that most matched target of current goal, in the present embodiment, it is possible to according to current
Detection block obtains and the matched prediction block of current detection frame.
Specifically, calculating the friendship of the detection block and prediction block and than IOU value, according to the value of the IOU whether default
When in range, determine whether detection block matches with the band of position of the prediction block in the current image frame.Wherein, when described
When the value of IOU is in the preset range, illustrate that the detection block is matched with the band of position of the prediction block in current image frame,
Conversely, illustrating the detection block and prediction block in current image frame when the value of the IOU is not in the preset range
The band of position mismatches.
Mode two: Feature Correspondence Algorithm is utilized, first eigenvector is matched with second feature vector.
Specifically, the Feature Correspondence Algorithm is for calculating between the first eigenvector and the second feature vector
Euclidean distance, it is determining special with the first eigenvector and second when determining that the Euclidean distance is within the scope of pre-determined distance
Levy the corresponding prediction block of vector and detection block matching, when determining the Euclidean distance not within the scope of pre-determined distance, determining and institute
It states first eigenvector and the corresponding prediction block of second feature vector and detection block mismatches.
Wherein, the Euclidean distance within the scope of pre-determined distance is smaller, illustrate the first eigenvector and second feature to
It measures corresponding prediction block and detection block more matches.
To sum up, the embodiment of the present invention combines above two mode, will test frame and prediction block in the current image frame
The band of position carry out IOU calculating, when the value of the IOU is in the preset range, the position of the detection block and prediction block
Set Region Matching success, while by the corresponding second feature vector of the detection block and it is preceding setting number of images frame in prediction block
Corresponding first eigenvector carries out Euclidean distance calculating by the sequence of the preceding setting number of images frame, when the Europe one by one
When formula distance is within the scope of pre-determined distance, the first eigenvector and the corresponding prediction block of second feature vector and detection block
With success.The picture frame of setting quantity before setting number of images frame as current image frame before described.
Step 13: use correcting mode corresponding with current motion state, using the prediction block to the detection block into
Row correction.
Specifically, correcting mode mainly includes following two mode:
Mode one:
1) it determines that current motion state is non-uniform variable motion state, determines corresponding first power of non-uniform variable motion state
Weight and the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
2) parameter of the prediction block is weighted using first weight, using the second weight to the inspection
The parameter for surveying frame is weighted, the detection block after being corrected.
In which one, since the current motion state of target is non-uniform variable motion state, illustrate the fortune of target at this time
Dynamic state is probably mutated, but prediction block is the previous frame according to current image frame to target in current image frame
Position predicted, therefore when being corrected to detection block, the results of more detection target positions according to detection blocks.
Therefore, the second weight and accounting for for the first weight are greater than preset value than difference again in the embodiment of the present invention, for example, described
Second weight accounts for again than being 0.9, and the first weight accounts for again than being 0.1, and the preset value is 0.49-0.79, then utilizes the first weight
The parameter of 0.1 pair of prediction block is weighted, and is weighted fortune to the parameter of the detection block using the second weight 0.9
It calculates, the detection block after being corrected, more testing results for believing detection block, even if can mutate in motion state
In the case of, it still is able to obtain effective tracking result.
Mode two:
1) when determining that current motion state is uniform variable motion state, corresponding first weight of uniform variable motion state is determined
With the second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first power
Weight accounts for again than difference no more than preset value;
2) parameter of the prediction block is weighted using first weight, using the second weight to the inspection
The parameter for surveying frame is weighted, the detection block after being corrected.
In which two, current motion state is uniform variable motion state, illustrates that prediction block passes through current image frame
When a upper picture frame is corrected current detection block, the future position that prediction block obtains is believable, the rail of target
Mark is in estimation range, and the position of target that the position of target that predicts of prediction block and detection block detect is in default model
In enclosing, therefore, the second weight and accounting for for the first weight are not more than preset value, also, the second weight and the first power than difference again
Weight accounts for again within a preset range than difference, at this point, being weighted fortune to the parameter of the prediction block using first weight
It calculates, is weighted using parameter of second weight to the detection block, the detection block after being corrected can be effective right
The target of the uniform variable motion carries out target tracking.
To sum up, method provided in an embodiment of the present invention, on the one hand can to current motion state be even speed change target into
On the other hand row target tracking can carry out target tracking to the target that current motion state is non-even speed change, compare existing skill
Art, application range is wider, and tracking effect is more preferable.
As shown in Fig. 2, being illustrated in conjunction with specific embodiments to the specific method step of target tracking:
Step 21: obtaining current image frame using monocular cam;
Step 22: being based on Kalman filtering algorithm, obtain the corresponding prediction block of target image in the current image frame
And detection block;
Step 23: by preparatory trained neural network model, extract target in the prediction frame region first is special
It levies vector and detects the second feature vector of target in frame region;
Step 24: calculate the friendship of the detection block and prediction block and than IOU value and the first eigenvector with it is described
Euclidean distance between second feature vector;
Step 25: whether within a preset range to judge the IOU value, it is no to then follow the steps 27 if it is execution step 26;
Step 26: determining that current motion state is uniform variable motion state, execute step 31;
Step 27: whether within a preset range judging the Euclidean distance, if so, executing step 28, otherwise execute step
Rapid 29;
Step 28: determining that current motion state is non-uniform variable motion state, execute step 30;
Step 29: initializing or abandon the prediction block;
Step 30: the parameter of the prediction block being weighted using first weight, utilizes the second weight pair
The parameter of the detection block is weighted, the detection block after being corrected, wherein the second weight accounts for weight with the first weight
It is greater than preset value than difference;
Step 31: the parameter of the prediction block being weighted using first weight, utilizes the second weight pair
The parameter of the detection block is weighted, the detection block after being corrected, wherein the second weight accounts for weight with the first weight
It is not more than preset value than difference.
Embodiment two
Based on identical inventive concept, the embodiment of the present invention two additionally provides a kind of equipment of target tracking, since this sets
Standby is the equipment in method in the embodiment of the present invention, and the principle that the equipment solves the problems, such as is similar to this method, therefore
The implementation of the equipment may refer to the implementation of method, and overlaps will not be repeated.
As shown in figure 3, the equipment includes: processor 300 and memory 301, wherein the memory 301 is stored with
Program code, when said program code is executed by the processor 300, so that the processor 300 is for executing following step
It is rapid:
Obtain current image frame and the corresponding prediction block of target and detection block in the current image frame;
By preparatory trained neural network model, extract in the prediction frame region first eigenvector of target and
Detect the second feature vector of target in frame region;
It will test frame to be matched with the band of position of the prediction block in the current image frame, and by the fisrt feature
Vector is matched with second feature vector, determines current motion state according to matching result;
Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block.
As an alternative embodiment, the processor 300 is specifically used for:
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state
With the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the processor 300 is specifically also used to:
Determine current motion state be uniform variable motion state when, determine corresponding first weight of uniform variable motion state and
Second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first weight
Account for again than difference no more than preset value;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the processor 300 is specifically used for:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the fisrt feature to
When amount is with secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, current motion state is determined
For uniform variable motion state.
As an alternative embodiment, the processor 300 is specifically used for:
Using Hungary matching algorithm, the band of position progress of frame and prediction block in the current image frame will test
Match;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
As an alternative embodiment, the processor 300 is specifically used for:
Calculate the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, really
Determine detection block to match with the band of position of the prediction block in the current image frame.
As an alternative embodiment, the processor 300 is specifically used for:
The Feature Correspondence Algorithm is European between the first eigenvector and the second feature vector for calculating
Distance, when determining that the Euclidean distance is within the scope of pre-determined distance, determining and the first eigenvector and second feature vector
Corresponding prediction block and detection block matching.
As an alternative embodiment, the processor 300 is specifically used for:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
As an alternative embodiment, the processor 300 is specifically used for:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection in the current image frame are obtained
Frame.
Embodiment three
Based on identical inventive concept, the embodiment of the present invention three additionally provides the equipment of another target tracking, due to this
Equipment is the equipment in the method in the embodiment of the present invention, and the principle that the equipment solves the problems, such as is similar to this method, because
The implementation of this equipment may refer to the implementation of method, and overlaps will not be repeated.
As shown in figure 4, the equipment includes: to obtain module 400, extraction module 401, matching module 402 and correction module
403, in which:
Obtain module 400, for obtain the corresponding prediction block of target in current image frame and the current image frame and
Detection block;
Extraction module 401, for extracting target in the prediction frame region by preparatory trained neural network model
First eigenvector and detection frame region in target second feature vector;
Matching module 402 is matched for will test frame with the band of position of the prediction block in the current image frame,
And match the first eigenvector with second feature vector, current motion state is determined according to matching result;
Correction module 403, for using correcting mode corresponding with current motion state, using the prediction block to described
Detection block is corrected.
As an alternative embodiment, the correction module 403 is specifically used for:
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state
With the second weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the equipment further includes the second correction module, it is used for:
Determine current motion state be uniform variable motion state when, determine corresponding first weight of uniform variable motion state and
Second weight, wherein the second weight accounts for again within a preset range than difference with the first weight, and the second weight and the first weight
Account for again than difference no more than preset value;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection
The parameter of frame is weighted, the detection block after being corrected.
As an alternative embodiment, the matching module 402 is specifically used for:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the fisrt feature to
When amount is with secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, current motion state is determined
For uniform variable motion state.
As an alternative embodiment, the matching module 402 is specifically used for:
Using Hungary matching algorithm, the band of position progress of frame and prediction block in the current image frame will test
Match;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
As an alternative embodiment, the matching module 402 is specifically used for:
Calculate the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, really
Determine detection block to match with the band of position of the prediction block in the current image frame.
As an alternative embodiment, the matching module 402 is specifically used for:
The Feature Correspondence Algorithm is European between the first eigenvector and the second feature vector for calculating
Distance, when determining that the Euclidean distance is within the scope of pre-determined distance, determining and the first eigenvector and second feature vector
Corresponding prediction block and detection block matching.
As an alternative embodiment, the acquisition module 400 is specifically used for:
According to the previous frame image of current image frame, to the centre coordinate of the target in previous frame picture frame, width and
Height is predicted, generates prediction block according to prediction result;
Current image frame is detected, according to centre coordinate, the width of the target in the obtained current image frame
And height generates detection block.
As an alternative embodiment, the acquisition module 400 is specifically used for:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection in the current image frame are obtained
Frame.
Example IV
The present invention provides a kind of computer storage medium, is stored thereon with computer program, which is executed by processor
Shi Shixian following steps:
Obtain current image frame and the corresponding prediction block of target and detection block in the current image frame;
By preparatory trained neural network model, extract in the prediction frame region first eigenvector of target and
Detect the second feature vector of target in frame region;
It will test frame to be matched with the band of position of the prediction block in the current image frame, and by the fisrt feature
Vector is matched with second feature vector, determines current motion state according to matching result;
Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The shape for the computer program product implemented in usable storage medium (including but not limited to magnetic disk storage and optical memory etc.)
Formula.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The equipment for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of equipment, the commander equipment realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (10)
1. a kind of method of target tracking, which is characterized in that this method comprises:
Obtain current image frame and the corresponding prediction block of target and detection block in the current image frame;
By preparatory trained neural network model, the first eigenvector of target and detection in the prediction frame region are extracted
The second feature vector of target in frame region;
It will test frame to be matched with the band of position of the prediction block in the current image frame, and by the first eigenvector
It is matched with second feature vector, current motion state is determined according to matching result;
Using correcting mode corresponding with current motion state, the detection block is corrected using the prediction block, wherein
It determines that current motion state is non-uniform variable motion state, determines corresponding first weight of non-uniform variable motion state and the second power
Weight, wherein the second weight and accounting for for the first weight are greater than preset value than difference again;Using first weight to the prediction block
Parameter be weighted, be weighted using parameter of second weight to the detection block, the inspection after being corrected
Survey frame.
2. the method according to claim 1, wherein further include:
When determining that current motion state is uniform variable motion state, corresponding first weight of uniform variable motion state and second are determined
Weight, wherein the second weight and the first weight account for again within a preset range than difference, and the second weight and the first weight account for
It is not more than preset value than difference again;
The parameter of the prediction block is weighted using first weight, using the second weight to the detection block
Parameter is weighted, the detection block after being corrected.
3. any method according to claim 1 ~ 2, which is characterized in that current motion state is determined according to matching result,
Include:
Determine that the band of position in the current image frame of detection block and prediction block mismatches, and the first eigenvector with
When secondary features vector matching, determine that current motion state is non-uniform variable motion state;
When determining that detection block is matched with the band of position of the prediction block in the current image frame, determine that current motion state is even
Variable motion state.
4. the method according to claim 1, wherein will test frame and prediction block in the current image frame
The band of position is matched, and the first eigenvector is matched with second feature vector, comprising:
Using Hungary matching algorithm, it will test frame and matched with the band of position of the prediction block in the current image frame;
Using Feature Correspondence Algorithm, first eigenvector is matched with second feature vector.
5. according to the method described in claim 4, it is characterized in that, will test frame and prediction block using Hungary matching algorithm
The band of position in the current image frame is matched, comprising:
It calculates the friendship of the detection block and prediction block and than IOU value, when according to the value of the IOU within a preset range, determines inspection
Frame is surveyed to match with the band of position of the prediction block in the current image frame.
6. according to the method described in claim 4, it is characterized in that, using Feature Correspondence Algorithm, by first eigenvector and
Two feature vectors are matched, comprising:
The Feature Correspondence Algorithm is used to calculate the Euclidean distance between the first eigenvector and the second feature vector,
When determining that the Euclidean distance is within the scope of pre-determined distance, determination is corresponding with the first eigenvector and second feature vector
Prediction block and detection block matching.
7. the method according to claim 1, wherein the target image obtained in the current image frame is corresponding
Prediction block and detection block, comprising:
According to the previous frame image of current image frame, to the centre coordinate, width and height of the target in previous frame picture frame
It is predicted, prediction block is generated according to prediction result;
Current image frame is detected, according to the centre coordinate of the target in the obtained current image frame, width and
Height generates detection block.
8. the method according to claim 1, wherein the target image obtained in the current image frame is corresponding
Prediction block and detection block, comprising:
Based on Kalman filtering algorithm, the corresponding prediction block of target image and the detection block in the current image frame are obtained.
9. a kind of equipment of target tracking, which is characterized in that the equipment includes: processor and memory, wherein the storage
Device is stored with program code, when said program code is executed by the processor, so that the processor perform claim requires 1
The step of ~ 8 any the method.
10. a kind of computer storage medium, is stored thereon with computer program, which is characterized in that the program is executed by processor
The step of Shi Shixian such as claim 1 ~ 8 any the method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910365924.4A CN109859239B (en) | 2019-05-05 | 2019-05-05 | A kind of method and apparatus of target tracking |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910365924.4A CN109859239B (en) | 2019-05-05 | 2019-05-05 | A kind of method and apparatus of target tracking |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109859239A CN109859239A (en) | 2019-06-07 |
CN109859239B true CN109859239B (en) | 2019-07-19 |
Family
ID=66889387
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910365924.4A Expired - Fee Related CN109859239B (en) | 2019-05-05 | 2019-05-05 | A kind of method and apparatus of target tracking |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109859239B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110298867B (en) * | 2019-06-21 | 2023-05-02 | 江西洪都航空工业集团有限责任公司 | Video target tracking method |
CN110929668A (en) * | 2019-11-29 | 2020-03-27 | 珠海大横琴科技发展有限公司 | Commodity detection method and device based on unmanned goods shelf |
CN111179311B (en) * | 2019-12-23 | 2022-08-19 | 全球能源互联网研究院有限公司 | Multi-target tracking method and device and electronic equipment |
CN111401194B (en) * | 2020-03-10 | 2023-09-22 | 北京百度网讯科技有限公司 | Data processing method and device for automatic driving vehicle |
CN111768427B (en) * | 2020-05-07 | 2023-12-26 | 普联国际有限公司 | Multi-moving-object tracking method, device and storage medium |
CN111982296B (en) * | 2020-08-07 | 2021-09-28 | 中国农业大学 | Moving target body surface temperature rapid detection method and system based on thermal infrared video |
CN112528925B (en) * | 2020-12-21 | 2024-05-07 | 深圳云天励飞技术股份有限公司 | Pedestrian tracking and image matching method and related equipment |
CN112528932B (en) * | 2020-12-22 | 2023-12-08 | 阿波罗智联(北京)科技有限公司 | Method and device for optimizing position information, road side equipment and cloud control platform |
CN116052062B (en) * | 2023-03-07 | 2023-06-16 | 深圳爱莫科技有限公司 | Robust tobacco display image processing method and device |
CN116563769B (en) * | 2023-07-07 | 2023-10-20 | 南昌工程学院 | Video target identification tracking method, system, computer and storage medium |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8295548B2 (en) * | 2009-06-22 | 2012-10-23 | The Johns Hopkins University | Systems and methods for remote tagging and tracking of objects using hyperspectral video sensors |
CN105070098B (en) * | 2015-07-14 | 2017-07-14 | 安徽清新互联信息科技有限公司 | A kind of vehicle distance detecting method based on car plate position |
CN108053427B (en) * | 2017-10-31 | 2021-12-14 | 深圳大学 | Improved multi-target tracking method, system and device based on KCF and Kalman |
-
2019
- 2019-05-05 CN CN201910365924.4A patent/CN109859239B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN109859239A (en) | 2019-06-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109859239B (en) | A kind of method and apparatus of target tracking | |
CN110349213B (en) | Pose determining method and device based on depth information, medium and electronic equipment | |
CN105405154B (en) | Target object tracking based on color-structure feature | |
CN110111363A (en) | A kind of tracking and equipment based on target detection | |
CN107452015B (en) | Target tracking system with re-detection mechanism | |
JP2016099941A (en) | System and program for estimating position of object | |
CN104517275A (en) | Object detection method and system | |
US10645668B2 (en) | Indoor positioning system and method based on geomagnetic signals in combination with computer vision | |
CN105374050A (en) | Moving target tracking recovery method and device | |
US20150161435A1 (en) | Frontal face detection apparatus and method using facial pose | |
CN110021034A (en) | A kind of tracking recording broadcasting method and system based on head and shoulder detection | |
CN106846367B (en) | A kind of Mobile object detection method of the complicated dynamic scene based on kinematic constraint optical flow method | |
CN112184757A (en) | Method and device for determining motion trail, storage medium and electronic device | |
CN108876758A (en) | Face identification method, apparatus and system | |
CN116883458B (en) | Transformer-based multi-target tracking system fusing motion characteristics with observation as center | |
CN109934849A (en) | Online multi-object tracking method based on track metric learning | |
CN110702093B (en) | Particle filter-based positioning method and device, storage medium and robot | |
Zhang et al. | NaviGlass: Indoor Localisation Using Smart Glasses. | |
Yi et al. | Ucmctrack: Multi-object tracking with uniform camera motion compensation | |
US8351653B2 (en) | Distance estimation from image motion for moving obstacle detection | |
CN103996207A (en) | Object tracking method | |
Mittal et al. | Pedestrian detection and tracking using deformable part models and Kalman filtering | |
CN107665495B (en) | Object tracking method and object tracking device | |
CN113192110A (en) | Multi-target tracking method, device, equipment and storage medium | |
CN105761504B (en) | Speed method for real-time measurement based on the collection of non-homogeneous video frame image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190719 |