CN109285348A - A kind of vehicle behavior recognition methods and system based on two-way length memory network in short-term - Google Patents
A kind of vehicle behavior recognition methods and system based on two-way length memory network in short-term Download PDFInfo
- Publication number
- CN109285348A CN109285348A CN201811261110.8A CN201811261110A CN109285348A CN 109285348 A CN109285348 A CN 109285348A CN 201811261110 A CN201811261110 A CN 201811261110A CN 109285348 A CN109285348 A CN 109285348A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- vehicle behavior
- model
- term
- video data
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/04—Detecting movement of traffic to be counted or controlled using optical or ultrasonic detectors
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle behavior recognition methods and system based on two-way length memory network in short-term, method includes: to transfer offline collected traffic video data;Detection tracking is carried out to the vehicle in traffic video data, extracts vehicle driving trace;Feature extraction is carried out to pretreated vehicle driving trace, and establishes training dataset and test data set;Model training is carried out to training dataset using two-way long short-term memory recurrent neural network, generates vehicle behavior identification model;Test data set is input in vehicle behavior identification model and carries out accuracy evaluation;By online real time collecting to traffic video data be input to the vehicle behavior identification model after accuracy evaluation carry out vehicle behavior identification, export recognition result.The present invention is by the way that based on two-way length, memory network is established vehicle behavior identification model and identified to the vehicle driving trace in traffic video data in short-term, to judge the traveling behavior of vehicle, accuracy of identification is high.
Description
Technical field
The present invention relates to vehicle behavior identification technology fields, and in particular to a kind of vehicle based on two-way length memory network in short-term
Activity recognition method and system.
Background technique
Vehicle drive Activity recognition based on video is the basis for establishing and improving intelligent transportation system, to raising path link
Row safety improves traffic congestion and is of great significance.Traditional road camera position is fixed, monitoring range is small, can not be to vehicle
Tracing detection is carried out under the complicated global scene such as intersection, with the fast development of unmanned plane and sensor integration technology,
The unmanned plane for carrying high-definition camera provides the collecting method of New Century Planned Textbook for field of traffic.Unmanned plane under large scene is handed over
Intervisibility frequency contains complicated and diversified vehicle behavior information, and track of vehicle is different in size, and track of vehicle Numerous uses tradition
Vehicle behavior analysis method under the single scene of camera, computationally intensive, prediction effect are also difficult to meet the requirements.
Therefore, the existing technology needs to be improved and developed.
Summary of the invention
The technical problem to be solved in the present invention is that in view of the above drawbacks of the prior art, providing a kind of based on two-way length
The vehicle behavior recognition methods of short-term memory network and system, it is intended to solve existing middle technology and use the single scene of traditional camera
Under vehicle behavior analysis method, computationally intensive, prediction effect is difficult to meet the requirements problem.
The technical proposal for solving the technical problem of the invention is as follows:
A kind of vehicle behavior recognition methods based on two-way length memory network in short-term, wherein the described method includes:
It transfers through unmanned plane that traffic route overhead is arranged in collected traffic video data offline;
The vehicle detection model established using depth convolutional neural networks carries out the vehicle in the traffic video data
Detection tracking, extracts vehicle driving trace;
The vehicle driving trace is pre-processed, feature extraction is carried out to pretreated vehicle driving trace, and
Training dataset and test data set are established according to the characteristic extracted;
Model training is carried out to the training dataset using two-way long short-term memory recurrent neural network, generates vehicle row
For identification model;
The test data set is input in vehicle behavior identification model and carries out accuracy evaluation;
By unmanned plane, collected traffic video data are input to the identification mould of the vehicle behavior after accuracy evaluation in real time
Type carries out the identification of vehicle behavior, exports recognition result.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein described to use depth convolution
The vehicle detection model of neural network carries out detection tracking to the vehicle in the traffic video data, extracts vehicle row
The step of sailing track specifically includes:
It, will offline collected traffic video data scaling processing after receiving offline collected traffic video data
After be input in multiple dimensioned depth convolutional neural networks and carry out model training, generate vehicle detection model;
Detection tracking is carried out to the vehicle in traffic video data using the vehicle detection model, extracts all vehicles
Driving trace.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein described to the vehicle row
It sails track and carries out pretreated step, specifically include:
The vehicle driving trace is smoothed, repeats point deletion processing and trace compression processing.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein the trace compression processing
It specifically includes:
The starting point of vehicle driving trace and terminating point are linked as straight line, and solved on the vehicle driving trace
The distance between all the points and the straight line;
Maximum distance is found out, and maximum distance is compared with preset threshold value;
When maximum distance is less than the threshold value, then the intermediate point on the vehicle driving trace is all deleted, and by institute
State approximation of the straight line as vehicle driving trace;
When maximum distance be not less than the threshold value, then retain tracing point corresponding to the maximum distance, and with the rail
Mark point is boundary, and vehicle driving trace is divided into two parts;
The step of trace compression processing is repeated to this two parts, until all maximum distance is all not more than the threshold
Value.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein described to pretreated
Vehicle driving trace carries out the step of feature extraction, specifically includes:
Tracing point angle value is extracted from pretreated vehicle driving trace.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein described to use two-way length
When memory recurrent neural network to the training dataset carry out model training, generate vehicle behavior identification model the step of, tool
Body includes:
By the hiding layer number in two-way long short-term memory recurrent neural network be set as 64, learning rate be set as 0.01 with
And the number of iterations is set as 10000;
Training dataset is input in two-way long short-term memory recurrent neural network and carries out model training, before then passing through
Predicted value is exported to propagation algorithm;
The error of predicted value and true value is calculated, and utilizes back-propagation algorithm by error back propagation;
Using the parameter in gradient decline principle more new model, and training is completed when minimizing objective function, generate vehicle
Activity recognition model.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein described by the test number
It is input in vehicle behavior identification model the step of carrying out accuracy evaluation according to collection, is specifically included:
Test data set is input in vehicle behavior model;
To be corresponded in recognition result and traffic video data that vehicle behavior identification model exports the traveling behavior of vehicle into
Row compares analysis, determines the precision of the vehicle behavior identification model.
The described vehicle behavior recognition methods based on two-way length memory network in short-term, wherein the determination vehicle
It the step of precision of Activity recognition model, specifically includes:
Vehicle will be corresponded in each class behavior result and traffic video data that vehicle behavior identification model is identified
Every a kind of traveling behavior be compared, export the precision of each class behavior;
The average value for calculating the precision of all class behaviors, as the final precision of vehicle behavior identification model.
A kind of vehicle behavior identifying system based on two-way length memory network in short-term, wherein the system comprises: setting exists
The unmanned plane in traffic route overhead, and the work station with the unmanned plane communication connection;The unmanned plane is used for traffic video
The acquisition of data, and collected data are sent to work station;
The work station includes:
Track obtains module, and the vehicle detection model for being established using depth convolutional neural networks is to the traffic video
Vehicle in data carries out detection tracking, extracts vehicle driving trace;
Characteristic extracting module, for being pre-processed to the vehicle driving trace, to pretreated vehicle driving rail
Mark carries out feature extraction, and the characteristic according to extraction establishes training dataset and test data set;
Model building module, for carrying out mould to the training dataset using two-way long short-term memory recurrent neural network
Type training generates vehicle behavior identification model;
Accuracy evaluation module is commented for the test data set to be input to progress precision in vehicle behavior identification model
Estimate;
Activity recognition module, for collected traffic video data to be input to after accuracy evaluation in real time by unmanned plane
Vehicle behavior identification model in carry out vehicle behavior identification, export recognition result.
The vehicle behavior identifying system based on two-way length memory network in short-term, wherein the accuracy evaluation module is also
For:
To be corresponded in recognition result and traffic video data that vehicle behavior identification model exports the traveling behavior of vehicle into
Row compares analysis, determines the precision of the vehicle behavior identification model.
Beneficial effects of the present invention: the present invention is by the way that based on two-way length, memory network is built using limited track sample in short-term
Vertical vehicle behavior identification model, identifies the vehicle driving trace in traffic video data using the model, to sentence
The traveling behavior of disconnected vehicle out, accuracy of identification are high.
Detailed description of the invention
Fig. 1 is the stream of the invention based on the preferred embodiment of the vehicle behavior recognition methods of memory network in short-term of two-way length
Cheng Tu.
Fig. 2 is two-way length in present invention memory network structural schematic diagram in short-term.
Fig. 3 is the principle frame of the invention based on work station in two-way length in short-term the vehicle behavior identifying system of memory network
Figure.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer and more explicit, right as follows in conjunction with drawings and embodiments
The present invention is further described.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and do not have to
It is of the invention in limiting.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Below
Description only actually at least one exemplary embodiment be it is illustrative, never as to the present invention and its application or make
Any restrictions.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
The present embodiment provides a kind of vehicle behavior recognition methods based on two-way length memory network in short-term, as shown in fig. 1,
The described method includes:
Step S100, it transfers through unmanned plane that traffic route overhead is arranged in collected traffic video data offline.
When it is implemented, the unmanned plane of the super clear camera of 4K is carried in the present embodiment in the biggish intersection overhead setting of the magnitude of traffic flow,
The unmanned plane is collected traffic video number for acquiring traffic video data, and including acquiring offline and online real time collecting
It according to later, is sent to the work station for having carried out communication connection in advance, so that work station carries out at data traffic video data
Reason.
Certainly, in the present embodiment, the quantity of unmanned plane setting can be set according to the actual conditions on traffic route
It sets, and the location layout of unmanned plane is also independently to be configured, to obtain comprehensive traffic video data.It is excellent
Selection of land, using unmanned plane, collected traffic video data carry out model training to the present embodiment offline.The traffic video data
In include the traffic intersection vehicle traveling information, such as the steering of vehicle.
Further, step S200, the traffic is regarded using the vehicle detection model that depth convolutional neural networks are established
Vehicle of the frequency in carries out detection tracking, extracts vehicle driving trace.
When it is implemented, after work station receives offline collected traffic video data, it will offline collected friendship
Multiple dimensioned depth convolutional neural networks (multi-scale convolutional is input to after logical video data calibration processing
Neural network, i.e. MSCNN) in carry out model training, generate vehicle detection model;Then the vehicle detection model is used
Detection tracking is carried out to the vehicle in traffic video data, extracts the driving trace of all vehicles.More rulers in the present embodiment
Depth convolutional neural networks are spent using traffic video data as training sample, and the study of multiple accesses is carried out to training sample,
The corresponding filter of each access carries out convolution operation, the characteristic dimension in traffic video data is obtained, then each
The characteristic dimension that access obtains is merged by a full articulamentum, so that a vehicle detection model is formed, vehicle inspection
Detection tracking can be carried out to the vehicle in traffic video data by surveying model, to detect the driving trace of each vehicle.
Vehicle detection model is constructed using multiple dimensioned depth convolutional neural networks in the present embodiment, vehicle can be got more accurately
Driving trace, reduce error.
Further, step S300, the vehicle driving trace is pre-processed, to pretreated vehicle driving rail
Mark carries out feature extraction, and the characteristic according to extraction establishes training dataset and test data set.
When it is implemented, driving trace is smoothed after work station gets vehicle driving trace,
Repeat the pretreatment operations such as point deletion processing and trace compression processing.Specifically, due to the testing result of vehicle detection model
There may be error, make occur noise in track of vehicle, it is therefore desirable to original driving trace data are smoothed, this
It is handled in implementation using sliding average smoothing method, calculation method are as follows:
Wherein, S is track of vehicle X, and Y sequence, N is sliding window size.
Also, since unmanned plane is the overhead that traffic intersection is arranged in, unmanned plane is using vertical view visual angle come to vehicle
What track extracted, and the halted states such as traffic lights are waited using overlooking the track of vehicle that visual angle is extracted and containing vehicle
Tracing point, these points for driving trace composition be repeat point.By comparing between the adjacent two o'clock in track in the present embodiment
Distance carries out repeating point deletion, calculation formula are as follows:
Wherein,
Δ indicates the distance between track front and back two o'clock, (xi+1, yi+1), (xi, yi) indicate adjacent track point.
Exist since vehicle detection model identifies the every frame of video, in the driving trace data extracted a large amount of superfluous
Remainder strong point, therefore track is compressed by using Douglas algorithm in the present embodiment, specific compression method are as follows:
1) starting point of vehicle driving trace and terminating point are linked as straight line, and solved on the vehicle driving trace
All the points and the distance between the straight line;
2) maximum distance D is found outmax, and by maximum distance DmaxIt is compared with preset threshold value T;
3) as maximum distance DmaxLess than the threshold value T, then the intermediate point on the vehicle driving trace is all deleted,
And using the straight line as the approximation of vehicle driving trace;
4) as maximum distance DmaxNot less than the threshold value T, then retain the maximum distance DmaxCorresponding tracing point, and
Using the tracing point as boundary, vehicle driving trace is divided into two parts;
5) step 1) of trace compression processing is repeated to 4), until all maximum distance is all little to this two parts
In the threshold value.
The threshold value T of this compression algorithm is bigger, and trace compression degree is bigger, and tracing point is deleted more, conversely, be retained
Tracing point is more, also more similar to former track, and constructed vehicle behavior identification model is also just more accurate in subsequent step.
After pre-processing to vehicle driving trace, the present embodiment carries out pretreated vehicle driving trace special
Sign is extracted, and is specifically, in the present embodiment the extraction tracing point difference value from pretreated vehicle driving trace, is then passed through
Tracing point coordinate difference score value calculates the angle change value of tracing point.It is specific as follows:
1) for the extraction of tracing point difference value, it is assumed that two neighboring tracing point li(xi, yi),li-1(xi-1, yi-1), then
The difference coordinate of i point is Δ li(xi-xi-1, yi-yi-1)。
2) for the extraction of tracing point angle value, such as liPoint and li-1The angle value that point is formed is just are as follows:
After feature extraction, training dataset and test data set are established according to the characteristic extracted in the present embodiment.
The training dataset constructs vehicle behavior identification model for being input to two-way long short-term memory recurrent neural network,
The test data set carries out accuracy evaluation for being input in the vehicle behavior identification model.
Further, step S400, the training dataset is carried out using two-way long short-term memory recurrent neural network
Model training generates vehicle behavior identification model.
Specifically, long short-term memory recurrent neural (LSTM) network carries out the information in cell state by door machine system
Management, including input gate, forgetting door and out gate.Door is forgotten to ht-1And xtIt is checked, to cell state ct-1In it is every
A value exports the number between one [0,1], and 0 indicates to abandon all information, and 1 indicates to retain all information, obtains ft, formula
It is as follows:
ft=σ (Wxfxt+Whfht-1+bf)
Input gate determines which new information needs to be retained in cell state, it is raw by sigmoid layers and tanh layers
At state renewal vector value gt, formula is as follows:
it=σ (Wxixt+Whiht-1+bi)
gt=tanh (Wxcxt+Whcht-1+bc)
A new cell state out, i.e. c can be updated by forgeing door and input gatet, formula is as follows:
ct=ft⊙ct-1+it⊙gt
The information for needing to export is determined finally by out gate, obtains output valve o by sigmoid layerst,
Its formula are as follows:
Ot=σ (WxoXt+Whoht-1+bo)
By cell state value by tanh, is exported multiplied by output layer, finally obtain the hidden state value h of output layert,
The value of information of output, formula are namely determined in the present embodiment are as follows:
ht=ot⊙tanh(ct)。
Due to the long short-term memory recurrent neural network of standard, track sets are only handled in chronological order, and the present embodiment
It is middle to use two-way long short-term memory recurrent neural network, as shown in Figure 2, in the present embodiment when carrying out vehicle behavior identification,
Each point complete past and following information in driving trace will be inputted and be supplied to output layer, captured with this more available
Information, by being introduced into the LSTM network (reversed LSTM layers in Fig. 2) of second layer extension standards, wherein hidden layer to hidden layer
Connection flowed with opposite time sequencing, the present embodiment has merged the defeated of the two directions by vector merging (feature connection)
Out, double output quantity is produced to next layer (the full articulamentum in Fig. 2), while in order to extract information relevant to classification,
The present embodiment is in hidden state ftIn be added to another output network (the output class label in Fig. 2), therefore make this reality
The model for applying example can be effectively combined information of the sequence in the past with future.
Specifically, when carrying out model training operation, training dataset is inputted first, then passes through propagated forward, according to
Above-mentioned principle calculates hidden layer sequence vector in two-way long short-term memory recurrent neural network, the hidden layer being arranged in the present embodiment
Quantity is 64, learning rate 0.01, the number of iterations 10000, and each propagated forward calculates output predicted value.Predicted value is calculated again
With the error of true value, and using back-propagation algorithm by error back propagation;Using in gradient decline principle more new model
Parameter, and be to complete training minimizing objective function, generate vehicle behavior identification model.
Preferably, since track data length is different in conventional model training process, usually using specified numerical value filling side
All trajectory processings are regular length by method, and this method treated data input network is trained, can be by all fillings
Numerical value be added calculate generate error, and be in the present embodiment by mark every track actual length, can training when dynamic
It determines effective length, ignores the invalid data of filling automatically, to obtain more accurate vehicle behavior identification model.
Further, the test data set step S500, is input to progress precision in vehicle behavior identification model to comment
Estimate.
An assessment is carried out in order to the accuracy of identification to vehicle behavior identification model, the test data in the present embodiment
Concentrating must be input to test data set in trained model carry out in this way comprising each class behavior sample for needing to identify
When identification, the recognition result of available each class behavior.Then, the present embodiment is first to each class behavior refinement degree, finally
The average value of the precision of all class behavior samples can be found out, and as final model accuracy.
Specifically, each class behavior result identified vehicle behavior identification model in the present embodiment and traffic regard
Every a kind of traveling behavior that frequency corresponds to vehicle in is compared, and exports the precision of each class behavior, then calculates institute
There is the average value of the precision of class behavior, as the final precision of vehicle behavior identification model.Specific calculation formula is as follows:
Wherein TPiCorrect quantity, FP are identified for the i-th classiFor the quantity of the i-th class identification mistake, PreiFor the knowledge of the i-th class
Other precision, Pre are mean accuracy value, the as final precision of vehicle behavior identification model.
Certainly, calculated final precision can be compared with preset precision threshold in the present embodiment, if finally
Progress is greater than preset precision threshold, then the vehicle behavior identification model meets required precision, can be used to the behavior to vehicle
It is identified.And if finally into preset precision threshold is less than, which is unsatisfactory for required precision, needs
The vehicle behavior identification model is continued to train, and repeats accuracy evaluation, until final precision is greater than preset essence
Until spending threshold value.
Further, step S600, by unmanned plane online real time collecting to traffic video data be input to by precision
Vehicle behavior identification model after assessment carries out the identification of vehicle behavior, exports recognition result.
After the completion of the building of vehicle behavior identification model and final precision is greater than precision threshold, and unmanned plane is adopted in real time online
Collect being input in the model of traffic video data, what vehicle behavior identification model will arrive unmanned plane online real time collecting
Vehicle driving trace is identified and is judged, so that it is determined which side steering the behavior of vehicle out, such as identification vehicle are toward, are straight
The behaviors such as row or lane change, and the behavior outcome output that will identify that.
By using two-way length, memory network identifies the collected vehicle behavior track of unmanned plane to the present invention in short-term,
Track sets past Future Information can be efficiently used, realizes the high-precision classification identification of vehicle behavior.The present invention being capable of dynamic
Track of vehicle length is determined to reduce error, efficient action learning model is established using limited track sample data, to not
Primary data predetermined speed is fast, and precision is high.
The present invention also provides a kind of vehicle behavior identifying system based on two-way length memory network in short-term, the system packets
It includes: the unmanned plane in traffic route overhead, and the work station with the unmanned plane communication connection is set;The unmanned plane is used for
The acquisition of traffic video data, and collected data are sent to work station.Specifically, as shown in Figure 3, in the present embodiment
Work station include:
Track obtains module 310, and the vehicle detection model for being established using depth convolutional neural networks is to the traffic
Vehicle in video data carries out detection tracking, extracts vehicle driving trace;
Characteristic extracting module 320, for being pre-processed to the vehicle driving trace, to pretreated vehicle driving
Track carries out feature extraction, and the characteristic according to extraction establishes training dataset and test data set;
Model building module 330, for using two-way long short-term memory recurrent neural network to the training dataset into
Row model training generates vehicle behavior identification model;
Accuracy evaluation module 340 carries out precision for the test data set to be input in vehicle behavior identification model
Assessment;
Activity recognition module 350, for collected traffic video data to be input to and comment by precision in real time by unmanned plane
The identification that vehicle behavior is carried out in vehicle behavior identification model after estimating, exports recognition result.
In the recognition result and traffic video data that accuracy evaluation module 350 is also used to export vehicle behavior identification model
The traveling behavior of corresponding vehicle is compared, and determines the precision of the vehicle behavior identification model.The work station it is each
The specific restriction of a module and functional effect are referring to above method embodiment.
In conclusion the present invention provides a kind of based on the vehicle behavior recognition methods of memory network in short-term of two-way length and is
System, method include: to transfer through unmanned plane that traffic route overhead is arranged in collected traffic video data offline;Use depth
The vehicle detection model that degree convolutional neural networks are established carries out detection tracking to the vehicle in the traffic video data, extracts
Vehicle driving trace;The vehicle driving trace is pre-processed, feature is carried out to pretreated vehicle driving trace and is mentioned
It takes, and the characteristic according to extraction establishes training dataset and test data set;Use two-way long short-term memory recurrent neural
Network carries out model training to the training dataset, generates vehicle behavior identification model;The test data set is input to
Accuracy evaluation is carried out in vehicle behavior identification model;By unmanned plane online real time collecting to traffic video data be input to by
Vehicle behavior identification model after accuracy evaluation carries out the identification of vehicle behavior, exports recognition result.The present invention passes through based on double
Vehicle behavior identification model is established to long memory network in short-term to identify to the vehicle driving trace in traffic video data,
To judge the traveling behavior of vehicle, accuracy of identification is high.
It should be understood that the application of the present invention is not limited to the above for those of ordinary skills can
With improvement or transformation based on the above description, all these modifications and variations all should belong to the guarantor of appended claims of the present invention
Protect range.
Claims (10)
1. a kind of vehicle behavior recognition methods based on two-way length memory network in short-term, which is characterized in that the described method includes:
It transfers through unmanned plane that traffic route overhead is arranged in collected traffic video data offline;
The vehicle detection model established using depth convolutional neural networks detects the vehicle in the traffic video data
Tracking, extracts vehicle driving trace;
The vehicle driving trace is pre-processed, feature extraction, and foundation are carried out to pretreated vehicle driving trace
The characteristic of extraction establishes training dataset and test data set;
Model training is carried out to the training dataset using two-way long short-term memory recurrent neural network, vehicle behavior is generated and knows
Other model;
The test data set is input in vehicle behavior identification model and carries out accuracy evaluation;
By unmanned plane online real time collecting to traffic video data be input to the vehicle behavior after accuracy evaluation identification mould
Type carries out the identification of vehicle behavior, exports recognition result.
2. the vehicle behavior recognition methods according to claim 1 based on two-way length memory network in short-term, which is characterized in that
The vehicle detection model established using depth convolutional neural networks detects the vehicle in the traffic video data
The step of tracking, extracting vehicle driving trace, specifically includes:
It, will be defeated after offline collected traffic video data scaling processing after receiving offline collected traffic video data
Enter into multiple dimensioned depth convolutional neural networks and carry out model training, generates vehicle detection model;
Detection tracking is carried out to the vehicle in traffic video data using the vehicle detection model, extracts the row of all vehicles
Sail track.
3. the vehicle behavior recognition methods according to claim 1 based on two-way length memory network in short-term, which is characterized in that
It is described that pretreated step is carried out to the vehicle driving trace, it specifically includes:
The vehicle driving trace is smoothed, repeats point deletion processing and trace compression processing.
4. the vehicle behavior recognition methods according to claim 3 based on two-way length memory network in short-term, which is characterized in that
The trace compression processing specifically includes:
The starting point of vehicle driving trace and terminating point are linked as straight line, and solved all on the vehicle driving trace
The distance between point and the straight line;
Maximum distance is found out, and maximum distance is compared with preset threshold value;
When maximum distance is less than the threshold value, then by all deletions of the intermediate point on the vehicle driving trace, and will be described straight
Approximation of the line as vehicle driving trace;
When maximum distance be not less than the threshold value, then retain tracing point corresponding to the maximum distance, and with the tracing point
For boundary, vehicle driving trace is divided into two parts;
The step of trace compression processing is repeated to this two parts, until all maximum distance is all not more than the threshold value.
5. the vehicle behavior recognition methods according to claim 1 based on two-way length memory network in short-term, which is characterized in that
It described the step of feature extraction is carried out to pretreated vehicle driving trace, specifically includes:
Tracing point angle value is extracted from pretreated vehicle driving trace.
6. the vehicle behavior recognition methods according to claim 1 based on two-way length memory network in short-term, which is characterized in that
It is described that model training is carried out to the training dataset using two-way long short-term memory recurrent neural network, it generates vehicle behavior and knows
It the step of other model, specifically includes:
By the hiding layer number in two-way long short-term memory recurrent neural network be set as 64, learning rate be set as 0.01 and repeatedly
Generation number is set as 10000;
Training dataset is input in two-way long short-term memory recurrent neural network and carries out model training, then by preceding to biography
Broadcast algorithm output predicted value;
The error of predicted value and true value is calculated, and utilizes back-propagation algorithm by error back propagation;
Using the parameter in gradient decline principle more new model, and training is completed when minimizing objective function, generate vehicle row
For identification model.
7. the vehicle behavior recognition methods according to claim 1 based on two-way length memory network in short-term, feature exist
In, it is described that the test data set is input in vehicle behavior identification model the step of carrying out accuracy evaluation, it specifically includes:
Test data set is input in vehicle behavior model;
The traveling behavior that vehicle is corresponded in recognition result and traffic video data that vehicle behavior identification model exports is compared
To analysis, the precision of the vehicle behavior identification model is determined.
8. the vehicle behavior recognition methods according to claim 7 based on two-way length memory network in short-term, feature exist
In, the precision of the determination vehicle behavior identification model the step of, specifically include:
The every of vehicle will be corresponded in each class behavior result and traffic video data that vehicle behavior identification model is identified
A kind of traveling behavior is compared, and exports the precision of each class behavior;
The average value for calculating the precision of all class behaviors, as the final precision of vehicle behavior identification model.
9. a kind of vehicle behavior identifying system based on two-way length memory network in short-term, which is characterized in that the system comprises: it sets
Set the unmanned plane in traffic route overhead, and the work station with the unmanned plane communication connection;The unmanned plane is used for traffic
The acquisition of video data, and collected data are sent to work station;
The work station includes:
Track obtains module, and the vehicle detection model for being established using depth convolutional neural networks is to the traffic video data
In vehicle carry out detection tracking, extract vehicle driving trace;
Characteristic extracting module, for being pre-processed to the vehicle driving trace, to pretreated vehicle driving trace into
Row feature extraction, and the characteristic according to extraction establishes training dataset and test data set;
Model building module, for carrying out model instruction to the training dataset using two-way long short-term memory recurrent neural network
Practice, generates vehicle behavior identification model;
Accuracy evaluation module carries out accuracy evaluation for the test data set to be input in vehicle behavior identification model;
Activity recognition module, for collected traffic video data to be input to the vehicle after accuracy evaluation in real time by unmanned plane
The identification that vehicle behavior is carried out in Activity recognition model, exports recognition result.
10. the vehicle behavior identifying system based on two-way length memory network in short-term according to claim 9, feature exist
In the accuracy evaluation module is also used to:
The traveling behavior that vehicle is corresponded in recognition result and traffic video data that vehicle behavior identification model exports is compared
To analysis, the precision of the vehicle behavior identification model is determined.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811261110.8A CN109285348B (en) | 2018-10-26 | 2018-10-26 | Vehicle behavior identification method and system based on unmanned aerial vehicle and long-and-short memory network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811261110.8A CN109285348B (en) | 2018-10-26 | 2018-10-26 | Vehicle behavior identification method and system based on unmanned aerial vehicle and long-and-short memory network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109285348A true CN109285348A (en) | 2019-01-29 |
CN109285348B CN109285348B (en) | 2022-02-18 |
Family
ID=65178018
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811261110.8A Active CN109285348B (en) | 2018-10-26 | 2018-10-26 | Vehicle behavior identification method and system based on unmanned aerial vehicle and long-and-short memory network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109285348B (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109739245A (en) * | 2019-02-19 | 2019-05-10 | 东软睿驰汽车技术(沈阳)有限公司 | One kind being based on unpiloted end to end model appraisal procedure and device |
CN109910909A (en) * | 2019-02-25 | 2019-06-21 | 清华大学 | A kind of interactive prediction technique of vehicle track net connection of more vehicle motion states |
CN110414375A (en) * | 2019-07-08 | 2019-11-05 | 北京国卫星通科技有限公司 | Recognition methods, device, storage medium and the electronic equipment of low target |
CN110796856A (en) * | 2019-10-16 | 2020-02-14 | 腾讯科技(深圳)有限公司 | Vehicle lane change intention prediction method and training method of lane change intention prediction network |
CN111079590A (en) * | 2019-12-04 | 2020-04-28 | 东北大学 | Peripheral vehicle behavior pre-judging method of unmanned vehicle |
CN111310583A (en) * | 2020-01-19 | 2020-06-19 | 中国科学院重庆绿色智能技术研究院 | Vehicle abnormal behavior identification method based on improved long-term and short-term memory network |
CN111723835A (en) * | 2019-03-21 | 2020-09-29 | 北京嘀嘀无限科技发展有限公司 | Vehicle movement track distinguishing method and device and electronic equipment |
CN112347993A (en) * | 2020-11-30 | 2021-02-09 | 吉林大学 | Expressway vehicle behavior and track prediction method based on vehicle-unmanned aerial vehicle cooperation |
WO2021109318A1 (en) * | 2019-12-03 | 2021-06-10 | 东南大学 | Method for estimating and predicting short-term traffic circulation state of urban road network |
CN112948715A (en) * | 2021-03-02 | 2021-06-11 | 杭州电子科技大学 | Vehicle classification method based on short-time GPS track data |
CN113065691A (en) * | 2021-03-22 | 2021-07-02 | 中国联合网络通信集团有限公司 | Traffic behavior prediction method and system |
CN113096379A (en) * | 2021-03-03 | 2021-07-09 | 东南大学 | Driving style identification method based on traffic conflict |
CN113327461A (en) * | 2021-08-03 | 2021-08-31 | 杭州海康威视数字技术股份有限公司 | Cooperative unmanned aerial vehicle detection method, device and equipment |
CN113511204A (en) * | 2020-03-27 | 2021-10-19 | 华为技术有限公司 | Vehicle lane changing behavior identification method and related equipment |
CN113674525A (en) * | 2021-07-30 | 2021-11-19 | 长安大学 | Signalized intersection vehicle queuing length prediction method based on sparse data |
CN115359662A (en) * | 2022-10-18 | 2022-11-18 | 智道网联科技(北京)有限公司 | Lane congestion prediction method and device |
CN117333847A (en) * | 2023-12-01 | 2024-01-02 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108305275A (en) * | 2017-08-25 | 2018-07-20 | 深圳市腾讯计算机***有限公司 | Active tracking method, apparatus and system |
-
2018
- 2018-10-26 CN CN201811261110.8A patent/CN109285348B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108305275A (en) * | 2017-08-25 | 2018-07-20 | 深圳市腾讯计算机***有限公司 | Active tracking method, apparatus and system |
Non-Patent Citations (1)
Title |
---|
JIASONG ZHU: "Bidirectional Long Short-Term Memory Network for Vehicle Behavior Recognition", 《WEB OF SCIENCE》 * |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109739245A (en) * | 2019-02-19 | 2019-05-10 | 东软睿驰汽车技术(沈阳)有限公司 | One kind being based on unpiloted end to end model appraisal procedure and device |
CN109910909B (en) * | 2019-02-25 | 2020-09-11 | 清华大学 | Automobile track internet interactive prediction method for multi-automobile motion state |
CN109910909A (en) * | 2019-02-25 | 2019-06-21 | 清华大学 | A kind of interactive prediction technique of vehicle track net connection of more vehicle motion states |
CN111723835A (en) * | 2019-03-21 | 2020-09-29 | 北京嘀嘀无限科技发展有限公司 | Vehicle movement track distinguishing method and device and electronic equipment |
CN110414375A (en) * | 2019-07-08 | 2019-11-05 | 北京国卫星通科技有限公司 | Recognition methods, device, storage medium and the electronic equipment of low target |
CN110796856A (en) * | 2019-10-16 | 2020-02-14 | 腾讯科技(深圳)有限公司 | Vehicle lane change intention prediction method and training method of lane change intention prediction network |
CN110796856B (en) * | 2019-10-16 | 2022-03-25 | 腾讯科技(深圳)有限公司 | Vehicle lane change intention prediction method and training method of lane change intention prediction network |
WO2021109318A1 (en) * | 2019-12-03 | 2021-06-10 | 东南大学 | Method for estimating and predicting short-term traffic circulation state of urban road network |
CN111079590A (en) * | 2019-12-04 | 2020-04-28 | 东北大学 | Peripheral vehicle behavior pre-judging method of unmanned vehicle |
CN111079590B (en) * | 2019-12-04 | 2023-05-26 | 东北大学 | Peripheral vehicle behavior pre-judging method of unmanned vehicle |
CN111310583A (en) * | 2020-01-19 | 2020-06-19 | 中国科学院重庆绿色智能技术研究院 | Vehicle abnormal behavior identification method based on improved long-term and short-term memory network |
CN111310583B (en) * | 2020-01-19 | 2023-02-10 | 中国科学院重庆绿色智能技术研究院 | Vehicle abnormal behavior identification method based on improved long-term and short-term memory network |
CN113511204A (en) * | 2020-03-27 | 2021-10-19 | 华为技术有限公司 | Vehicle lane changing behavior identification method and related equipment |
CN112347993A (en) * | 2020-11-30 | 2021-02-09 | 吉林大学 | Expressway vehicle behavior and track prediction method based on vehicle-unmanned aerial vehicle cooperation |
CN112948715A (en) * | 2021-03-02 | 2021-06-11 | 杭州电子科技大学 | Vehicle classification method based on short-time GPS track data |
CN113096379A (en) * | 2021-03-03 | 2021-07-09 | 东南大学 | Driving style identification method based on traffic conflict |
CN113065691A (en) * | 2021-03-22 | 2021-07-02 | 中国联合网络通信集团有限公司 | Traffic behavior prediction method and system |
CN113674525A (en) * | 2021-07-30 | 2021-11-19 | 长安大学 | Signalized intersection vehicle queuing length prediction method based on sparse data |
CN113327461B (en) * | 2021-08-03 | 2021-11-23 | 杭州海康威视数字技术股份有限公司 | Cooperative unmanned aerial vehicle detection method, device and equipment |
CN113327461A (en) * | 2021-08-03 | 2021-08-31 | 杭州海康威视数字技术股份有限公司 | Cooperative unmanned aerial vehicle detection method, device and equipment |
CN115359662A (en) * | 2022-10-18 | 2022-11-18 | 智道网联科技(北京)有限公司 | Lane congestion prediction method and device |
CN115359662B (en) * | 2022-10-18 | 2023-01-10 | 智道网联科技(北京)有限公司 | Lane congestion prediction method and device |
CN117333847A (en) * | 2023-12-01 | 2024-01-02 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
CN117333847B (en) * | 2023-12-01 | 2024-03-15 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
Also Published As
Publication number | Publication date |
---|---|
CN109285348B (en) | 2022-02-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109285348A (en) | A kind of vehicle behavior recognition methods and system based on two-way length memory network in short-term | |
CN104615983B (en) | Activity recognition method based on recurrent neural network and human skeleton motion sequence | |
CN109448370B (en) | Traffic control subarea division method based on vehicle track data | |
CN108319972B (en) | End-to-end difference network learning method for image semantic segmentation | |
CN105069415B (en) | Method for detecting lane lines and device | |
CN103235933B (en) | A kind of vehicle abnormality behavioral value method based on HMM | |
CN109460709A (en) | The method of RTG dysopia analyte detection based on the fusion of RGB and D information | |
CN108171112A (en) | Vehicle identification and tracking based on convolutional neural networks | |
CN105005999A (en) | Obstacle detection method for blind guiding instrument based on computer stereo vision | |
CN113139470B (en) | Glass identification method based on Transformer | |
CN107038713A (en) | A kind of moving target method for catching for merging optical flow method and neutral net | |
CN110032952B (en) | Road boundary point detection method based on deep learning | |
CN105046717A (en) | Robust video object tracking method | |
CN107564290A (en) | A kind of urban road intersection saturation volume rate computational methods | |
CN103003846A (en) | Articulation region display device, articulation region detection device, articulation region relatedness computation device, articulation shape region relatedness computation device, and articulation region display method | |
CN108320051B (en) | Mobile robot dynamic collision avoidance planning method based on GRU network model | |
CN115147790B (en) | Future track prediction method of vehicle based on graph neural network | |
CN115690153A (en) | Intelligent agent track prediction method and system | |
CN106599810A (en) | Head pose estimation method based on stacked auto-encoding | |
CN115457277A (en) | Intelligent pavement disease identification and detection method and system | |
CN113903173B (en) | Vehicle track feature extraction method based on directed graph structure and LSTM | |
Shao et al. | Failure detection for motion prediction of autonomous driving: An uncertainty perspective | |
CN113076988B (en) | Mobile robot vision SLAM key frame self-adaptive screening method based on neural network | |
CN112364890A (en) | Intersection guiding method for making urban navigable network by taxi track | |
CN115392407B (en) | Non-supervised learning-based danger source early warning method, device, equipment and medium |
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 |