CN107492113A - A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method - Google Patents
A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method Download PDFInfo
- Publication number
- CN107492113A CN107492113A CN201710402241.2A CN201710402241A CN107492113A CN 107492113 A CN107492113 A CN 107492113A CN 201710402241 A CN201710402241 A CN 201710402241A CN 107492113 A CN107492113 A CN 107492113A
- Authority
- CN
- China
- Prior art keywords
- target
- moving
- sequence
- video
- model
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The present invention relates to a kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method, training moving target position forecast model needs to collect video under single fixed scene first, multiple target tracking is carried out to video, then generate the timing coordination sequence of each target, filter out the timing coordination arrangement set of single target class, real trace arrangement set is inputted into network model, output is compared repeatedly with True Data, it can obtain every kind of single classification target position prediction model, single goal class position prediction model prediction Future Positions or track corresponding to each target class are used in actual video.In practical application, usage scenario and Training scene need to be same or like.To sum up the present invention is that predicted motion target location and track provide a kind of new method in video image using deep learning method, and moving target can be including pedestrian, motor vehicle, non-motor vehicle etc..
Description
Technical field
The present invention relates to deep learning and art of image analysis, more particularly to a kind of moving object in video sequences position
Forecast model training method, position predicting method and trajectory predictions method.
Background technology
Moving target position prediction in video image refers in video image by the existing fortune of each moving target
Dynamic rail mark, their position coordinateses at the following specified moment are predicted.Movement objective orbit in video image is predicted
Refer to by the existing movement locus of each moving target in video image, to each of which, following track is predicted.Gu
Determine the trajectory predictions of the moving target under scene or position prediction can apply to the friendship of some outdoor scenes such as crossroad
Logical monitoring and control, the passenger flow that can also be applied to some indoor scenes such as station are dredged, and look-ahead goes out in video image
Wagon flow people's flow distribution, can predict it can happen that, so as to facilitate the relevant personnel to carry out countermeasure in advance, if
It was found that the actual stream of people or wagon flow track（Or position）Excessive with the trajector deviation of previous prediction, that illustrates that this block region may be sent out
Some abnormal conditions are given birth to, the relevant personnel can pay close attention to immediately.
A kind of pixel of moving object in video sequences is disclosed in the A of Chinese invention patent specification CN 105913454
Grid Track Forecasting Methodology, this method is according to the history pixel trace information that moving target is obtained from video image, it is proposed that
A kind of fractional model fitting locus formula based on camera image-forming principle, establishing over-determined systems with reference to historical track can be in the hope of
The fractional model coefficient, so that it is determined that linear uniform motion target pixel coordinate and time in video image in real space
Between relation, the pixel coordinate of final Accurate Prediction future time instance target.
At present, it can be used for practical grind for what moving object in video sequences trajectory predictions and moving target position were predicted
Study carefully achievement and few, but for reality, similar to the motion state of moving target as pedestrian, non-motor vehicle and motor vehicle
All there is certain inheritance and continuity, i.e., following state is always based on current known trajectory, and every kind of motion mesh
Mark has different motion features, such as speed, radius of turn etc., learns the true of different motion target using deep learning method
Real movement locus is feasible so as to predict the method for the track of different types of moving target or position.
The content of the invention
The technical problem to be solved in the present invention is:A kind of moving object in video sequences position prediction model training is provided
Method, and it is pre- to provide a kind of a kind of position for training to obtain based on moving object in video sequences position prediction model training method
The position coordinates surveyed after model prediction moving target specified time, and one kind is provided and is based on a kind of moving object in video sequences position
Put forecast model training method and train obtained position prediction model prediction moving target motion rail following in video image
Mark, moving target can include pedestrian, motor vehicle, non-motor vehicle etc..
In order to solve the above technical problems, the present invention takes following technical scheme：
It is pre- that the present invention provides a kind of moving object in video sequences position prediction model training method, position predicting method and track
Survey method, wherein, a kind of moving object in video sequences position prediction model training method, comprise the following steps：
（1）Under single fixed scene, collect a number of video image, and on each image, by multiple target with
Track, automatic marking go out the position of each moving target central point and the identity information of each moving target of generation；
（2）By sample frequency f frames/second, the coordinate of each moving target with identity information in extracting per two field picture is corresponding every
The moving target of individual different identity information generates a respective timing coordination sequence, thus forms timing coordination arrangement set；
（3）According to the targeted species for being actually needed detection, filtered out in the timing coordination arrangement set generated from step 2 and meet bar
The target class of part, and the sequence that length is less than sequential neural network model input data length L is picked out, thus form the target
The timing coordination arrangement set of class；
Wherein, the sequential neural network model be training used in network model, refer to the net that can handle time series data
Network model, include the deformation of single or multiple lift RNN neutral nets or single or multiple lift RNN neutral nets, the deformation of RNN networks
Including LSTM, GRU etc., the network model is preferably LSTM network models, more preferably multilayer LSTM network models；
（4）Generate the input trajectory list X and actual comparison value list Y of neural network model：The timing coordination generated from step 3
Extracted successively in a certain bar sequence in arrangement setThe continuous sequential coordinate data deposit X lists in common L of position, its
In, if this sequence length is, then, it is necessary to the position after the predicted motion target n seconds, then k=
F*n,And be integer, while extract thePosition data deposit Y lists, are finished until this sequence extracts,
Timing coordination arrangement set is extracted one by one in the same manner, until every sequence in timing coordination arrangement set has extracted
Finish；
（5）Inputted list X as neural network model, list Y uses depth as actual value, and according to target generic
Learning method is spent, predicted position is compared repeatedly with physical location, adjusts the ginseng of the target class movement position forecast model
Number, finally train the position prediction model of the target class；
（6）Repeat stepIt can train to obtain different target class position prediction models, if only needing to detect a kind of target,
The step need not be performed.
K=1 described in the step 4 described in the above method, the moving target position forecast model training method train to obtain
Model can be used for predicted motion target continuous position, i.e. movement objective orbit, the moving target position forecast model can also
Referred to as movement objective orbit forecast model.
The present invention also provides a kind of moving object in video sequences position predicting method, based on video image described above
The method of middle moving target position prediction, the moving target position Forecasting Methodology comprise the following steps:
（1）With under the same or analogous scene of Training scene, by with identical sample frequency f frames/second during training pattern, read
One-frame video data, by multiple target tracking, it is every to obtain the position of each moving target central point and generation in the two field picture
The identity information of individual moving target；
（2）The moving target of corresponding different identity information correspondingly generates respective timing coordination sequence, if the sequential of moving target
Coordinate sequence has been present, then need not generate again, the positional information of moving target in the two field picture is added to corresponding
Timing coordination sequence in；
（3）The input data length L of network used in being more than and train when the length of timing coordination sequence corresponding to some moving target,
According to the species of prediction target, train what is obtained to a kind of above-mentioned moving object in video sequences position prediction model training method
The newest L bit timing coordinate sequences of corresponding target class position prediction mode input, model output result are target after the n seconds
Position coordinates；
(4) repeat step 3 can obtain the predicted position of other moving targets in video.
Wherein, n described in the step 3 and a kind of above-mentioned moving object in video sequences position prediction model training method
N in the step 4 is same numerical value.
The present invention also provides a kind of moving object in video sequences trajectory predictions method, based on video image described above
The situation of k in the method for middle moving target position prediction=1, obtained forecast model also known as be movement objective orbit forecast model, general
Step 3 and 4 replaces with following steps 3 and 4 in a kind of above-mentioned moving object in video sequences position predicting method：
（3）When the length of timing coordination sequence corresponding to some moving target is more than the input data length of neutral net used in training
L is spent, according to the species of prediction target, is trained to a kind of above-mentioned moving object in video sequences position prediction model training method
The newest L bit timing coordinate sequences of the corresponding target class trajectory predictions mode input that arrives, model output result are target under
The coordinate at one moment, the coordinate predicted is added in the timing coordination sequence of the target, it is defeated to trajectory predictions model again
Enter newest L bit timing coordinate sequences, model output result is coordinate of the target in subsequent time, so circulation, can be somebody's turn to do
Position coordinates of the target at continuous more moment is predicted, that is, obtains the trajectory predictions of the target, wherein each moment isSecond；
（4）Repeat step 3 can obtain the prediction locus of other moving targets in video.
The present invention inputs network mould by gathering video under fixed scene, by true movement locus known to regular length
Type, the predicted value that network exports is compared repeatedly with actual value, constantly adjust out optimal single goal type games trajectory predictions ginseng
Exponential model and single goal type games position prediction parameter model, single goal class corresponding to each target class is used in actual video
Single goal movement position Prediction Parameters mould corresponding to movement locus Prediction Parameters model prediction track and/or each target class of use
The position coordinates of moving target after type predicts certain moment, the present invention are the predicted motion in video image using deep learning method
Target location and track provide a kind of new method, and moving target can include pedestrian, motor vehicle, non-motor vehicle etc..
Brief description of the drawings
Fig. 1 is a kind of moving object in video sequences trajectory predictions model training method provided in an embodiment of the present invention and fortune
Moving-target position prediction model training method flow chart.
Fig. 2 is training movement objective orbit model provided in an embodiment of the present invention with making in training moving target position model
Dict1, Dict2 and single goal class timing coordination arrangement set exemplary plot.
Fig. 3 is a kind of moving object in video sequences trajectory predictions method flow diagram provided in an embodiment of the present invention.
Fig. 4 is a kind of moving object in video sequences position predicting method flow chart provided in an embodiment of the present invention.
Embodiment
Technical scheme is described in detail below in conjunction with the accompanying drawings.
Due to being directed to multiple target tracking in following specific embodiments, for convenience of understanding, multiple target tracking skill is described first
Art:
Multiple target tracking, i.e. Multiple Object Tracking (MOT), also referred to as Multiple Target Tracking
(MTT), its main task is in one group of image sequence is given, and finds the target moved in the image sequence, and by different frame
In moving target correspond, finally provide the movement locus of different target, these moving targets can be arbitrary, such as
Pedestrian, vehicle, sportsman, various animals etc..Existing target tracking algorism is divided into two major classes, and the first kind is primarily upon effect
Lifting, such as MDNet (Multi-Domain Network), TCNN is another kind of, compares concern tracking velocity, such as
Staple, GOTURN.
Due to the present invention innovative point do not lie in multi-object tracking method, so in the present invention multi-object tracking method choosing
Select the development based on actual Multitarget Tracking, hardware technology, the multi-object tracking method that the present invention uses at this stage can be with
One kind for more than in four kinds.
Fig. 1 is a kind of moving object in video sequences trajectory predictions model training method flow provided in an embodiment of the present invention
Figure, reference picture 1, the model training method mainly comprise the following steps:
1）Under single fixed scene, collect a number of video image, and on each image, by multiple target with
Track, automatic marking go out the position of each moving target central point and the identity information of each moving target of generation.
Wherein, the fixed scene includes indoor or outdoors scene, described to refer to training pattern under single fixed scene
Used video image is collected under a fixed scene, the time of video image shooting, such as day and night, no
It is construed as limiting；The moving target includes pedestrian, motor vehicle, non-motor vehicle.The identity information refers to identify each target
The ID of body, include the species and numbering of the target, such as Walker_1, wherein " walker " refers to pedestrian, " 1 " refers to the pedestrian's
Numbering.
Following stepFor generate each moving target in video image timing coordination sequence, it is necessary to explanation
It is that the wherein order of operations and/or operating method can be rearranged.The schematic diagram of reference picture 2 can be best understood from
Dict1, Dict2 and single goal class timing coordination arrangement set.
2）By certain sample frequency selecting video frame, according to the frame number at place, each identity is carried in extracting per two field picture
The coordinate of the moving target of information, Dict1 is generated, wherein, key assignments key is frame number, and data value corresponding to each key assignments is should
The identity information of each moving target and position coordinates this moment in frame.
Wherein, the sample frequency can select according to actual conditions, such as actual video is that 25 frames are per second, general one second
The change of clock video image is simultaneously little, can take a two field picture with per second, and now the sample frequency is 1 frame/second, for the ease of reason
Solve, sample frequency is appointed as 1 frame/second in all embodiments of the invention；Wherein, if multiple target tracking image sources in step 1
In single video, frame number described in the step is the frame number of the image in video, if image sources are in multiple videos, the step
The frame number can be that the frame number of the image in video adds timestamp, and main purpose is each in order to prevent from being subsequently drawn into
The track of moving target is chaotic, and certain timestamp can be other marks.
3）Dict1 data are extracted, generate Dict2, wherein, key is the target that each carries identity information, each
Value corresponding to key is the timing coordination sequence of this target.
Wherein, the timing coordination sequence of the target refers to the target according to the tactic coordinate sequence of frame number, sequential
Refer to the order according to time order and function.
4）According to the targeted species for being actually needed detection, filtered out respectively from Dict2 according to identity information eligible
Target class, and pick out length be less than sequential neural network model input data length L sequence, thus form the target class
Timing coordination arrangement set.
Wherein, the sequential neural network model refers to that the neural network model for analyzing time series data can be used for, including
The deformation of single or multiple lift RNN (recurrent neural network) neutral nets or single or multiple lift RNN neutral nets,
The deformation of RNN networks is including LSTM, GRU etc., because LSTM network models training effect is other better than RNN models and RNN models
Deformation, therefore preferably LSTM network models, because the training effect of multilayer LSTM network models is better than individual layer, therefore further preferably
For multilayer LSTM network models；Network model input data the length L, L can be 15,18,20 or 22 etc., according to training
The Detection results of the model come, select optimal L values.
5）Generate the input list X and actual comparison value list Y of multilayer LSTM networks：From step 4 formation sequence set
A certain bar sequence in extract successivelyThe continuous sequential coordinate data deposit X lists in common L of position, wherein, if this
Sequence length is, then, and extract theIndividual data are stored in Y lists, until this sequence is taken out
Take complete, extract arrangement set one by one in the same manner.
Wherein, to wall scroll sequence extraction process in a specific example, it is assumed that the length of certain timing coordination sequence is
23, network model input length is 20, then is extracted successivelyPosition,Position andPosition deposit X lists, extract the 21st
Position, 22 and 23 deposit Y lists.
6）Using list X as multilayer LSTM network inputs, list Y uses as actual value, and according to target generic
Deep learning method, predicted position is compared repeatedly with physical location, adjusts the target class movement locus forecast model
Parameter, finally train the trajectory predictions model of the target class.
Wherein, the trajectory predictions model of the target class can not be trained once to optimal, repeat step sometimesIt can increase
New video image carrys out Optimized model performance, and the sample frequency of the trajectory predictions model of the target class in training refers to step
The sample frequency in 2.
7）Repeat stepIt can train to obtain different target class trajectory predictions models, if only needing to detect a kind of mesh
Mark, then the step need not be performed.
Fig. 3 is a kind of moving object in video sequences trajectory predictions method flow diagram provided in an embodiment of the present invention, reference
Fig. 3, the trajectory predictions method mainly comprise the following steps:
1）With under the same or analogous scene of Training scene, being read in real time by identical sample frequency during training trajectory predictions model
Camera video data are taken, by multiple target tracking, obtain the position of each moving target central point and life in the two field picture
Into the identity information of each moving target.
Wherein, the similar scene refers to similar to the scene for gathering training video data, is in particular in two scenes
The movement law of lower moving target is similar；Wherein, the sample frequency is identical with sample frequency during training forecast model, described
Sample frequency is 1 frame/second in the present embodiment.
2）The moving target of corresponding different identity information correspondingly generates respective timing coordination sequence（If moving target when
Sequence coordinate sequence has been present, then need not generate again）, by the positional information of moving target in the two field picture be added to pair
In the timing coordination sequence answered.
3）When the length of timing coordination sequence corresponding to some moving target is more than the input data length of network used in training
Spend L, then can be according to the targeted species for wanting prediction locus, the locus model input of the corresponding target class drawn to training is newest
L bit timing coordinate sequences, predict coordinate of the target in subsequent time image, using the coordinate predicted as predict it is next
The coordinate predicted, i.e., be added in the timing coordination sequence corresponding to the target, so by the part input of frame coordinate first
Newest L bit timings coordinate sequence input trajectory forecast model is predicted into coordinate of the target in subsequent time image afterwards,
So circulation, the coordinate that can obtain the target at the continuous moment is predicted, that is, obtains the prediction locus of the target.
Wherein, the subsequent time is related to sample frequency during trajectory predictions model training, and frequency is sampled in the present embodiment
Rate is 1 frame/second, then subsequent time is that, if sample frequency is f, subsequent time corresponds to after 1 secondAfter second, each moment pair
It should beSecond.
4）Repeat step 3 can obtain the prediction locus of different motion target.
Fig. 1 is a kind of moving object in video sequences position prediction model training method flow provided in an embodiment of the present invention
Figure, reference picture 1, the model training method mainly comprise the following steps:
It should be noted that：StepWith above-described embodiment -- a kind of moving object in video sequences trajectory predictions model training
Method is identical.
1）Under single fixed scene, a number of video image is collected, and on each image, pass through multiple target
Tracking, automatic marking go out the position of each moving target central point and the identity information of each moving target of generation.
Following stepFor generating the timing coordination sequence of each moving target in video image, it is necessary to illustrate,
The wherein order of operations and/or operating method can be rearranged.The schematic diagram of reference picture 2 can be best understood from
Dict1, Dict2 and single goal class timing coordination arrangement set.
2）By certain sample frequency selecting video frame, according to the frame number at place, each identity is carried in extracting per two field picture
The coordinate of the moving target of information, Dict1 is generated, wherein, key assignments key is frame number, and data value corresponding to each key assignments is should
The identity information of each moving target and position coordinates this moment in frame.
Wherein, sample frequency is appointed as 1 frame/second in the present embodiment.
3）Dict1 data are extracted, generate Dict2, wherein, key is the target that each carries identity information, each
Value corresponding to key is the timing coordination sequence of this target.
4）According to the targeted species for being actually needed detection, filtered out respectively from Dict2 according to identity information eligible
Target class, and pick out length be less than multilayer LSTM network model input data length L sequence, thus form the target class
Timing coordination arrangement set；
5）Generate the input list X and actual comparison value list Y of multilayer LSTM networks：One from step 4 formation sequence set
Extracted successively in bar sequenceThe continuous sequential coordinate data deposit X row in common L of position
Table, wherein set this sequence length as, and extract theIndividual data are stored in Y lists, until this sequence extracts
Finish, extract arrangement set one by one in the same manner.
Wherein, the number at the time of k is wants last time at intervals in prediction time and input timing, definition sampling frequency
Rate is f frames/second, then each moment correspond toSecond, in Y lists some data and the sequence in corresponding X lists it is last when
Carve intervalSecond, sample frequency is 1 frame/second in the present embodiment, then each moment is 1 second, if k access value 40, in Y lists
The last moment interval 40 seconds of some data and the sequence in corresponding X lists, k selection directly affects the position that training obtains
The prediction result of forecast model, show as position prediction model predictionThe coordinate of moving target in the picture after second, above-mentioned reality
Apply example --- a kind of moving object in video sequences trajectory predictions model training method is actually the situation of k=1.The k is most
The forecast model that whole value comes out according to the complexity of actual scene, sample frequency, the coverage of video pictures, hands-on
Effect, the factor such as user's request determine, in general, if this factor of the complexity of actual scene is only considered, for letter
Single scene, k can be with value, for complex scene, k can be with value。
Wherein, to wall scroll sequence extraction process in a specific example, it is assumed that the length of certain timing coordination sequence is
64, network model input length is 20, k 40, then is extracted successivelyPosition,Position,Position,With
X lists are stored in, extract the 60th, 61,62,63 and 64 deposit Y list.
6）Using list X as multilayer LSTM network inputs, list Y uses as actual value, and according to target generic
Deep learning method, predicted position is compared repeatedly with physical location, adjusts the parameter of the target class position prediction model,
Finally train the position prediction model of the target class.
Wherein, the position prediction model of the target class can not be trained once to optimal, repeat step sometimesIt can increase
New video image carrys out Optimized model performance.
7）Repeat stepIt can train to obtain different target class position prediction models, if only needing to detect a kind of target,
The step need not so be performed.
Fig. 4 is a kind of moving object in video sequences position predicting method flow chart provided in an embodiment of the present invention, reference
Fig. 4, the position predicting method mainly comprise the following steps:
It should be noted that：StepWith above-described embodiment -- a kind of moving object in video sequences trajectory predictions method phase
Together.
1）With under the same or analogous scene of Training scene, it is real by identical sample frequency during training trajectory predictions model
When read camera video data, by multiple target tracking, obtain the position of each moving target central point in the two field picture with
And the identity information of each moving target of generation.
2）The moving target of corresponding different identity information correspondingly generates respective timing coordination sequence（If moving target when
Sequence coordinate sequence has been present, then need not generate again）, by the positional information of moving target in the two field picture be added to pair
In the timing coordination sequence answered.
3）When the length of timing coordination sequence corresponding to some moving target is more than the input data length of network used in training
Spend L, then can be according to the targeted species for wanting predicted position, the position prediction mode input of the corresponding target class drawn to training
L bit timing coordinate sequences, the coordinate of the target in the picture after predicting k moment.
Wherein, the number at the time of k is wants last time at intervals in prediction time and input timing, if sample frequency
For f, each moment corresponds toSecond, i.e., after position forecast model can predict the k/f seconds, the position of moving target in the picture.
4）Repeat step 3 can obtain the predicted position of other moving targets.
The present invention inputs network mould by gathering video under fixed scene, by true movement locus known to regular length
Type, the predicted value that network exports is compared repeatedly with actual value, constantly adjust out optimal single goal type games trajectory predictions ginseng
Exponential model and single goal type games position prediction parameter model, single goal class corresponding to each target class is used in actual video
Single goal movement position Prediction Parameters mould corresponding to movement locus Prediction Parameters model prediction track and/or each target class of use
The position coordinates of moving target after type predicts certain moment.The present invention is in practical application, actual use scene and Training scene need
It is same or like, video is collected in advance for different usage scenarios and is trained, and can obtain preferable prediction effect.To sum up
The present invention provides a kind of new method for the predicted motion target trajectory in video image and predicted motion target location, and this method can
With the moving target being each detected in prognostic chart picture, moving target can include pedestrian, motor vehicle, non-motor vehicle etc..
Claims (6)
- A kind of 1. moving object in video sequences position prediction model training method, it is characterised in that：Comprise the following steps：（1）Under single fixed scene, collect a number of video image, and on each image, by multiple target with Track, automatic marking go out the position of each moving target central point and the identity information of each moving target of generation；（2）By sample frequency f frames/second, the coordinate of each moving target with identity information in extracting per two field picture is corresponding every The moving target of individual different identity information generates a respective timing coordination sequence, thus forms timing coordination arrangement set；（3）According to the targeted species for being actually needed detection, filtered out in the timing coordination arrangement set generated from step 2 and meet bar The target class of part, and the sequence that length is less than sequential neural network model input data length L is picked out, thus form the target The timing coordination arrangement set of class；（4）Generate the input trajectory list X and actual comparison value list Y of sequential neural network model：The sequential generated from step 3 Extracted successively in a certain bar sequence in coordinate sequence setThe continuous sequential coordinate data deposit X row in common L of position Table, wherein, if this sequence length is, then, it is necessary to the position after the predicted motion target n seconds, Then k=f*n,And be integer, while extract thePosition data deposit Y lists, until this sequence extracts Finish, timing coordination arrangement set is extracted one by one in the same manner, until every sequence in timing coordination arrangement set Extraction finishes；（5）Inputted list X as sequential neural network model, list Y makes as actual value, and according to target generic With deep learning method, predicted position is compared repeatedly with physical location, adjusts the target class movement position forecast model Parameter, finally train the position prediction model of the target class；（6）Repeat stepIt can train to obtain different target class position prediction models, if only needing to detect a kind of target, The step need not be performed.
- 2. a kind of moving object in video sequences position prediction model training method according to claim 1, its feature exist In the sequential neural network model is the deformation of single or multiple lift RNN network models, single or multiple lift RNN network models.
- 3. a kind of moving object in video sequences position prediction model training method according to claim 1, its feature exist In the sequential neural network model is multilayer LSTM network models.
- 4. a kind of moving object in video sequences position prediction model training method according to claim 1,2 or 3, it is special Sign is, described in a kind of moving object in video sequences position prediction model training method described in claim 1,2 or 3 K=1 described in step 4, the model that the moving target position forecast model training method trains to obtain can be used in predicted motion mesh Mark track.
- 5. a kind of moving object in video sequences position predicting method, it is characterised in that based on described in claim any one of 1-3 A kind of moving object in video sequences position prediction model training method, the moving object in video sequences position prediction side Method comprises the following steps：（1）With under the same or analogous scene of Training scene, by with identical sample frequency f frames/second during training pattern, read One-frame video data, by multiple target tracking, it is every to obtain the position of each moving target central point and generation in the two field picture The identity information of individual moving target；（2）The moving target of corresponding different identity information correspondingly generates respective timing coordination sequence, if the sequential of moving target Coordinate sequence has been present, then need not generate again, the positional information of moving target in the two field picture is added to corresponding Timing coordination sequence in；（3）The input data length L of network used in being more than and train when the length of timing coordination sequence corresponding to some moving target, According to the species of prediction target, the corresponding mesh for training to obtain to any one of claim 1-3 movement objective orbit training methods The newest L bit timing coordinate sequences of class position prediction mode input are marked, model output result is that position of the target after the n seconds is sat Mark；（4）Repeat step 3 can obtain the predicted position of other moving targets in video.
- A kind of 6. moving object in video sequences trajectory predictions method, it is characterised in that regarded based on one kind described in claim 4 Moving target position forecast model training method trains obtained moving target position forecast model in frequency image, by claim Step 3-4 replaces with following steps 3 and 4 described in a kind of moving object in video sequences position predicting method described in 5：（3）The input data length L of network used in being more than and train when the length of timing coordination sequence corresponding to some moving target, According to the species of prediction target, the corresponding target class position prediction mode input for training to obtain to claim 4 methods described Newest L bit timing coordinate sequences, model output result be target in the coordinate of subsequent time, the coordinate predicted is added to In the timing coordination sequence of the target, again to the newest L bit timing coordinate sequences of position prediction mode input, model output knot Fruit is coordinate of the target in subsequent time, so circulation, can obtain position coordinates prediction of the target at continuous more moment, produce To the trajectory predictions of the target, wherein each moment isSecond；（4）Repeat step 3 can obtain the prediction locus of other moving targets in video.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710402241.2A CN107492113B (en) | 2017-06-01 | 2017-06-01 | A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710402241.2A CN107492113B (en) | 2017-06-01 | 2017-06-01 | A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107492113A true CN107492113A (en) | 2017-12-19 |
| CN107492113B CN107492113B (en) | 2019-11-05 |
Family
ID=60642091
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710402241.2A Active CN107492113B (en) | 2017-06-01 | 2017-06-01 | A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107492113B (en) |
Cited By (45)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053653A (en) * | 2018-01-11 | 2018-05-18 | 广东蔚海数问大数据科技有限公司 | Vehicle behavior Forecasting Methodology and device based on LSTM |
| CN108229407A (en) * | 2018-01-11 | 2018-06-29 | 武汉米人科技有限公司 | A kind of behavioral value method and system in video analysis |
| CN108257156A (en) * | 2018-01-24 | 2018-07-06 | 清华大学深圳研究生院 | A kind of method of the automatic tracing target object from video |
| CN108537278A (en) * | 2018-04-10 | 2018-09-14 | 中国人民解放军火箭军工程大学 | A kind of Multi-source Information Fusion single goal location determining method and system |
| CN108764560A (en) * | 2018-05-22 | 2018-11-06 | 电子科技大学 | Aircraft scene trajectory predictions method based on shot and long term Memory Neural Networks |
| CN108776974A (en) * | 2018-05-24 | 2018-11-09 | 南京行者易智能交通科技有限公司 | A kind of real-time modeling method method suitable for public transport scene |
| CN108960086A (en) * | 2018-06-20 | 2018-12-07 | 电子科技大学 | Based on the multi-pose human body target tracking method for generating confrontation network positive sample enhancing |
| CN109410496A (en) * | 2018-10-25 | 2019-03-01 | 北京交通大学 | Attack early warning method, apparatus and electronic equipment |
| CN109508812A (en) * | 2018-10-09 | 2019-03-22 | 南京航空航天大学 | A kind of aircraft Trajectory Prediction method based on profound memory network |
| CN109598066A (en) * | 2018-12-05 | 2019-04-09 | 百度在线网络技术（北京）有限公司 | Effect evaluation method, device, equipment and the storage medium of prediction module |
| CN109769285A (en) * | 2019-02-18 | 2019-05-17 | 西安交通大学 | A kind of method for routing of the inter-vehicular communication based on position prediction |
| CN109829439A (en) * | 2019-02-02 | 2019-05-31 | 京东方科技集团股份有限公司 | The calibration method and device of a kind of pair of head motion profile predicted value |
| CN109829936A (en) * | 2019-01-29 | 2019-05-31 | 青岛海信网络科技股份有限公司 | A kind of method and apparatus of target tracking |
| CN109902588A (en) * | 2019-01-29 | 2019-06-18 | 北京奇艺世纪科技有限公司 | A kind of gesture identification method, device and computer readable storage medium |
| CN109961045A (en) * | 2019-03-25 | 2019-07-02 | 联想(北京)有限公司 | A kind of location information prompt method, device and electronic equipment |
| CN110111358A (en) * | 2019-05-14 | 2019-08-09 | 西南交通大学 | A kind of method for tracking target based on multilayer temporal filtering |
| CN110135314A (en) * | 2019-05-07 | 2019-08-16 | 电子科技大学 | A kind of multi-object tracking method based on depth Trajectory prediction |
| CN110210417A (en) * | 2019-06-05 | 2019-09-06 | 深圳前海达闼云端智能科技有限公司 | Method, terminal and readable storage medium for predicting pedestrian motion trail |
| CN110248178A (en) * | 2019-06-18 | 2019-09-17 | 深圳大学 | Utilize the viewport prediction technique and system of object tracking and historical track panoramic video |
| CN110334654A (en) * | 2019-07-08 | 2019-10-15 | 北京地平线机器人技术研发有限公司 | Video estimation method and apparatus, the training method of video estimation model and vehicle |
| CN110488811A (en) * | 2019-07-22 | 2019-11-22 | 上海有个机器人有限公司 | A method of the robot based on social networks model predicts pedestrian track |
| CN110648531A (en) * | 2019-09-19 | 2020-01-03 | 军事科学院系统工程研究院网络信息研究所 | Node mobility prediction method based on deep learning in vehicle-mounted self-organizing network |
| CN110728698A (en) * | 2019-09-30 | 2020-01-24 | 天津大学 | Multi-target tracking model based on composite cyclic neural network system |
| CN110781710A (en) * | 2018-12-17 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Target object clustering method and device |
| CN110895876A (en) * | 2018-09-11 | 2020-03-20 | 深圳云天励飞技术有限公司 | Vehicle monitoring method and device and storage medium |
| CN110909625A (en) * | 2019-11-04 | 2020-03-24 | 向仲宇 | Computer vision basic network training, identifying and constructing method and device |
| CN111026115A (en) * | 2019-12-13 | 2020-04-17 | 华南智能机器人创新研究院 | Robot obstacle avoidance control method and device based on deep learning |
| CN111031351A (en) * | 2020-03-11 | 2020-04-17 | 北京三快在线科技有限公司 | Method and device for predicting target object track |
| CN111027461A (en) * | 2019-12-06 | 2020-04-17 | 长安大学 | Vehicle track prediction method based on multi-dimensional single-step LSTM network |
| CN111127510A (en) * | 2018-11-01 | 2020-05-08 | 杭州海康威视数字技术股份有限公司 | Target object position prediction method and device |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| CN111223121A (en) * | 2020-01-13 | 2020-06-02 | 清华大学 | Multi-target track prediction method based on collision-free LSTM |
| WO2020114136A1 (en) * | 2018-12-06 | 2020-06-11 | 西安光启未来技术研究院 | Method and device for evaluating algorithm performance |
| CN111310735A (en) * | 2020-03-18 | 2020-06-19 | 桂林电子科技大学 | Automobile track prediction method based on LSTM technology |
| CN111428763A (en) * | 2020-03-17 | 2020-07-17 | 陕西师范大学 | Pedestrian trajectory prediction method based on scene constraint GAN |
| CN111611703A (en) * | 2020-05-15 | 2020-09-01 | 深圳星地孪生科技有限公司 | Sand table deduction method, device, equipment and storage medium based on digital twins |
| CN111746557A (en) * | 2019-03-26 | 2020-10-09 | 通用汽车环球科技运作有限责任公司 | Path planning fusion for vehicles |
| WO2020224479A1 (en) * | 2019-05-06 | 2020-11-12 | 腾讯科技（深圳）有限公司 | Method and apparatus for acquiring positions of target, and computer device and storage medium |
| CN112466158A (en) * | 2020-11-26 | 2021-03-09 | 东南大学 | Plane intersection-oriented vehicle collision risk assessment and prediction method |
| CN112529940A (en) * | 2020-12-17 | 2021-03-19 | 北京深睿博联科技有限责任公司 | Moving target position prediction method and device under fixed camera |
| CN110310350B (en) * | 2019-06-24 | 2021-06-11 | 清华大学 | Animation-based motion prediction generation method and device |
| CN113129330A (en) * | 2020-01-14 | 2021-07-16 | 北京地平线机器人技术研发有限公司 | Track prediction method and device for movable equipment |
| CN113689671A (en) * | 2021-07-20 | 2021-11-23 | 广州市炬盾科技发展有限公司 | Urban underground space security alarm method and system |
| US11753005B2 (en) | 2019-12-23 | 2023-09-12 | Baidu International Technology (Shenzhen) Co., Ltd. | Collision detection method, and device, as well as electronic device and storage medium |
| CN117406170A (en) * | 2023-12-15 | 2024-01-16 | 中科华芯(东莞)科技有限公司 | Positioning method and system based on ultra-wideband |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010070128A1 (en) * | 2008-12-19 | 2010-06-24 | Thomson Licensing | Method for multi-resolution motion estimation |
| CN105335986A (en) * | 2015-09-10 | 2016-02-17 | 西安电子科技大学 | Characteristic matching and MeanShift algorithm-based target tracking method |
| CN105913454A (en) * | 2016-04-06 | 2016-08-31 | 东南大学 | Pixel coordinate locus prediction method of motion object in video image |
| CN106022239A (en) * | 2016-05-13 | 2016-10-12 | 电子科技大学 | Multi-target tracking method based on recurrent neural network |
| CN106204646A (en) * | 2016-07-01 | 2016-12-07 | 湖南源信光电科技有限公司 | Multiple mobile object tracking based on BP neutral net |
| CN106296739A (en) * | 2016-08-11 | 2017-01-04 | 清华大学苏州汽车研究院（吴江） | The moving target detecting method of a kind of improvement and system |
-
2017
- 2017-06-01 CN CN201710402241.2A patent/CN107492113B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010070128A1 (en) * | 2008-12-19 | 2010-06-24 | Thomson Licensing | Method for multi-resolution motion estimation |
| CN105335986A (en) * | 2015-09-10 | 2016-02-17 | 西安电子科技大学 | Characteristic matching and MeanShift algorithm-based target tracking method |
| CN105913454A (en) * | 2016-04-06 | 2016-08-31 | 东南大学 | Pixel coordinate locus prediction method of motion object in video image |
| CN106022239A (en) * | 2016-05-13 | 2016-10-12 | 电子科技大学 | Multi-target tracking method based on recurrent neural network |
| CN106204646A (en) * | 2016-07-01 | 2016-12-07 | 湖南源信光电科技有限公司 | Multiple mobile object tracking based on BP neutral net |
| CN106296739A (en) * | 2016-08-11 | 2017-01-04 | 清华大学苏州汽车研究院（吴江） | The moving target detecting method of a kind of improvement and system |
Non-Patent Citations (2)
| Title |
|---|
| DANIELE FERNANDES E SILVA 等: "A Levels-of-Precision Approach for Simulating Multiple Physics-Based Soft Tissues", 《2016 29TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI)》 * |
| 王长军: "基于视频的目标检测与跟踪技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (67)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053653B (en) * | 2018-01-11 | 2021-03-30 | 广东蔚海数问大数据科技有限公司 | Vehicle behavior prediction method and device based on LSTM |
| CN108229407A (en) * | 2018-01-11 | 2018-06-29 | 武汉米人科技有限公司 | A kind of behavioral value method and system in video analysis |
| CN108053653A (en) * | 2018-01-11 | 2018-05-18 | 广东蔚海数问大数据科技有限公司 | Vehicle behavior Forecasting Methodology and device based on LSTM |
| CN108257156A (en) * | 2018-01-24 | 2018-07-06 | 清华大学深圳研究生院 | A kind of method of the automatic tracing target object from video |
| CN108537278A (en) * | 2018-04-10 | 2018-09-14 | 中国人民解放军火箭军工程大学 | A kind of Multi-source Information Fusion single goal location determining method and system |
| CN108537278B (en) * | 2018-04-10 | 2019-07-16 | 中国人民解放军火箭军工程大学 | A kind of Multi-source Information Fusion single goal location determining method and system |
| CN108764560A (en) * | 2018-05-22 | 2018-11-06 | 电子科技大学 | Aircraft scene trajectory predictions method based on shot and long term Memory Neural Networks |
| CN108776974A (en) * | 2018-05-24 | 2018-11-09 | 南京行者易智能交通科技有限公司 | A kind of real-time modeling method method suitable for public transport scene |
| CN108960086A (en) * | 2018-06-20 | 2018-12-07 | 电子科技大学 | Based on the multi-pose human body target tracking method for generating confrontation network positive sample enhancing |
| CN108960086B (en) * | 2018-06-20 | 2021-06-04 | 电子科技大学 | Multi-pose human body target tracking method based on generation of confrontation network positive sample enhancement |
| CN110895876A (en) * | 2018-09-11 | 2020-03-20 | 深圳云天励飞技术有限公司 | Vehicle monitoring method and device and storage medium |
| CN109508812A (en) * | 2018-10-09 | 2019-03-22 | 南京航空航天大学 | A kind of aircraft Trajectory Prediction method based on profound memory network |
| CN109410496A (en) * | 2018-10-25 | 2019-03-01 | 北京交通大学 | Attack early warning method, apparatus and electronic equipment |
| CN111127510B (en) * | 2018-11-01 | 2023-10-27 | 杭州海康威视数字技术股份有限公司 | Target object position prediction method and device |
| CN111127510A (en) * | 2018-11-01 | 2020-05-08 | 杭州海康威视数字技术股份有限公司 | Target object position prediction method and device |
| CN109598066B (en) * | 2018-12-05 | 2023-08-08 | 百度在线网络技术（北京）有限公司 | Effect evaluation method, apparatus, device and storage medium for prediction module |
| CN109598066A (en) * | 2018-12-05 | 2019-04-09 | 百度在线网络技术（北京）有限公司 | Effect evaluation method, device, equipment and the storage medium of prediction module |
| WO2020114136A1 (en) * | 2018-12-06 | 2020-06-11 | 西安光启未来技术研究院 | Method and device for evaluating algorithm performance |
| CN110781710A (en) * | 2018-12-17 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Target object clustering method and device |
| CN109902588A (en) * | 2019-01-29 | 2019-06-18 | 北京奇艺世纪科技有限公司 | A kind of gesture identification method, device and computer readable storage medium |
| CN109829936B (en) * | 2019-01-29 | 2021-12-24 | 青岛海信网络科技股份有限公司 | Target tracking method and device |
| CN109829936A (en) * | 2019-01-29 | 2019-05-31 | 青岛海信网络科技股份有限公司 | A kind of method and apparatus of target tracking |
| CN109829439A (en) * | 2019-02-02 | 2019-05-31 | 京东方科技集团股份有限公司 | The calibration method and device of a kind of pair of head motion profile predicted value |
| CN109829439B (en) * | 2019-02-02 | 2020-12-29 | 京东方科技集团股份有限公司 | Method and device for calibrating predicted value of head motion track |
| CN109769285A (en) * | 2019-02-18 | 2019-05-17 | 西安交通大学 | A kind of method for routing of the inter-vehicular communication based on position prediction |
| CN109961045A (en) * | 2019-03-25 | 2019-07-02 | 联想(北京)有限公司 | A kind of location information prompt method, device and electronic equipment |
| CN109961045B (en) * | 2019-03-25 | 2021-10-22 | 联想(北京)有限公司 | Position information prompting method and device and electronic equipment |
| CN111746557A (en) * | 2019-03-26 | 2020-10-09 | 通用汽车环球科技运作有限责任公司 | Path planning fusion for vehicles |
| CN111746557B (en) * | 2019-03-26 | 2024-03-29 | 通用汽车环球科技运作有限责任公司 | Path plan fusion for vehicles |
| JP7154678B2 (en) | 2019-05-06 | 2022-10-18 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Target position acquisition method, device, computer equipment and computer program |
| WO2020224479A1 (en) * | 2019-05-06 | 2020-11-12 | 腾讯科技（深圳）有限公司 | Method and apparatus for acquiring positions of target, and computer device and storage medium |
| JP2022518745A (en) * | 2019-05-06 | 2022-03-16 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Target position acquisition method, equipment, computer equipment and computer program |
| CN110135314A (en) * | 2019-05-07 | 2019-08-16 | 电子科技大学 | A kind of multi-object tracking method based on depth Trajectory prediction |
| CN110111358A (en) * | 2019-05-14 | 2019-08-09 | 西南交通大学 | A kind of method for tracking target based on multilayer temporal filtering |
| CN110111358B (en) * | 2019-05-14 | 2022-05-24 | 西南交通大学 | Target tracking method based on multilayer time sequence filtering |
| CN110210417A (en) * | 2019-06-05 | 2019-09-06 | 深圳前海达闼云端智能科技有限公司 | Method, terminal and readable storage medium for predicting pedestrian motion trail |
| CN110248178B (en) * | 2019-06-18 | 2021-11-23 | 深圳大学 | Viewport prediction method and system using object tracking and historical track panoramic video |
| CN110248178A (en) * | 2019-06-18 | 2019-09-17 | 深圳大学 | Utilize the viewport prediction technique and system of object tracking and historical track panoramic video |
| CN110310350B (en) * | 2019-06-24 | 2021-06-11 | 清华大学 | Animation-based motion prediction generation method and device |
| CN110334654A (en) * | 2019-07-08 | 2019-10-15 | 北京地平线机器人技术研发有限公司 | Video estimation method and apparatus, the training method of video estimation model and vehicle |
| CN110488811B (en) * | 2019-07-22 | 2023-04-07 | 上海有个机器人有限公司 | Method for predicting pedestrian track by robot based on social network model |
| CN110488811A (en) * | 2019-07-22 | 2019-11-22 | 上海有个机器人有限公司 | A method of the robot based on social networks model predicts pedestrian track |
| CN110648531B (en) * | 2019-09-19 | 2020-12-04 | 军事科学院系统工程研究院网络信息研究所 | Node mobility prediction method based on deep learning in vehicle-mounted self-organizing network |
| CN110648531A (en) * | 2019-09-19 | 2020-01-03 | 军事科学院系统工程研究院网络信息研究所 | Node mobility prediction method based on deep learning in vehicle-mounted self-organizing network |
| CN110728698B (en) * | 2019-09-30 | 2023-05-16 | 天津大学 | Multi-target tracking system based on composite cyclic neural network system |
| CN110728698A (en) * | 2019-09-30 | 2020-01-24 | 天津大学 | Multi-target tracking model based on composite cyclic neural network system |
| CN110909625A (en) * | 2019-11-04 | 2020-03-24 | 向仲宇 | Computer vision basic network training, identifying and constructing method and device |
| CN111027461B (en) * | 2019-12-06 | 2022-04-29 | 长安大学 | Vehicle track prediction method based on multi-dimensional single-step LSTM network |
| CN111027461A (en) * | 2019-12-06 | 2020-04-17 | 长安大学 | Vehicle track prediction method based on multi-dimensional single-step LSTM network |
| CN111026115A (en) * | 2019-12-13 | 2020-04-17 | 华南智能机器人创新研究院 | Robot obstacle avoidance control method and device based on deep learning |
| CN111114554A (en) * | 2019-12-16 | 2020-05-08 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| CN111114554B (en) * | 2019-12-16 | 2021-06-11 | 苏州智加科技有限公司 | Method, device, terminal and storage medium for predicting travel track |
| US11753005B2 (en) | 2019-12-23 | 2023-09-12 | Baidu International Technology (Shenzhen) Co., Ltd. | Collision detection method, and device, as well as electronic device and storage medium |
| CN111223121A (en) * | 2020-01-13 | 2020-06-02 | 清华大学 | Multi-target track prediction method based on collision-free LSTM |
| CN113129330A (en) * | 2020-01-14 | 2021-07-16 | 北京地平线机器人技术研发有限公司 | Track prediction method and device for movable equipment |
| CN111031351A (en) * | 2020-03-11 | 2020-04-17 | 北京三快在线科技有限公司 | Method and device for predicting target object track |
| CN111428763B (en) * | 2020-03-17 | 2023-04-07 | 陕西师范大学 | Pedestrian trajectory prediction method based on scene constraint GAN |
| CN111428763A (en) * | 2020-03-17 | 2020-07-17 | 陕西师范大学 | Pedestrian trajectory prediction method based on scene constraint GAN |
| CN111310735A (en) * | 2020-03-18 | 2020-06-19 | 桂林电子科技大学 | Automobile track prediction method based on LSTM technology |
| CN111611703A (en) * | 2020-05-15 | 2020-09-01 | 深圳星地孪生科技有限公司 | Sand table deduction method, device, equipment and storage medium based on digital twins |
| CN112466158A (en) * | 2020-11-26 | 2021-03-09 | 东南大学 | Plane intersection-oriented vehicle collision risk assessment and prediction method |
| CN112466158B (en) * | 2020-11-26 | 2022-04-29 | 东南大学 | Plane intersection-oriented vehicle collision risk assessment and prediction method |
| CN112529940A (en) * | 2020-12-17 | 2021-03-19 | 北京深睿博联科技有限责任公司 | Moving target position prediction method and device under fixed camera |
| CN112529940B (en) * | 2020-12-17 | 2022-02-11 | 北京深睿博联科技有限责任公司 | Moving target position prediction method and device under fixed camera |
| CN113689671A (en) * | 2021-07-20 | 2021-11-23 | 广州市炬盾科技发展有限公司 | Urban underground space security alarm method and system |
| CN113689671B (en) * | 2021-07-20 | 2022-11-22 | 广州市炬盾科技发展有限公司 | Urban underground space security alarm method and system |
| CN117406170A (en) * | 2023-12-15 | 2024-01-16 | 中科华芯(东莞)科技有限公司 | Positioning method and system based on ultra-wideband |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107492113B (en) | 2019-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107492113B (en) | A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method | |
| Murugesan et al. | Efficient anomaly detection in surveillance videos based on multi layer perception recurrent neural network | |
| CN110378259A (en) | A kind of multiple target Activity recognition method and system towards monitor video | |
| CN108803617A (en) | Trajectory predictions method and device | |
| CN110490901A (en) | The pedestrian detection tracking of anti-attitudes vibration | |
| CN110363131B (en) | Abnormal behavior detection method, system and medium based on human skeleton | |
| RU2471231C1 (en) | Method to search for objects in sequence of images produced from stationary video camera | |
| CN108875603A (en) | Intelligent driving control method and device, electronic equipment based on lane line | |
| CN102750710B (en) | Moving target statistical method and device in a kind of image | |
| CN106169188A (en) | A kind of method for tracing object based on the search of Monte Carlo tree | |
| CN105208325B (en) | The land resources monitoring and early warning method captured and compare analysis is pinpointed based on image | |
| CN108791302B (en) | Driver behavior modeling system | |
| CN109583373B (en) | Pedestrian re-identification implementation method | |
| CN110047096B (en) | A kind of multi-object tracking method and system based on depth conditions random field models | |
| CN112241969A (en) | Target detection tracking method and device based on traffic monitoring video and storage medium | |
| JP6789876B2 (en) | Devices, programs and methods for tracking objects using pixel change processed images | |
| CN108920805B (en) | Driver behavior modeling system with state feature extraction function | |
| CN109448363A (en) | Suspected vehicles based on trajectory predictions intelligently sealed method, system and processing terminal | |
| CN103150552B (en) | A kind of driving training management method based on number of people counting | |
| Perše et al. | Analysis of multi-agent activity using petri nets | |
| CN108734109A (en) | A kind of visual target tracking method and system towards image sequence | |
| CN112241974A (en) | Traffic accident detection method, processing method, system and storage medium | |
| CN114372503A (en) | Cluster vehicle motion trail prediction method | |
| CN109286782A (en) | Multiple video camera makes an inspection tour the method and device of control automatically | |
| CN103605960B (en) | A kind of method for identifying traffic status merged based on different focal video 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 |