CN110516556A - Multi-target tracking detection method, device and storage medium based on Darkflow-DeepSort - Google Patents
Multi-target tracking detection method, device and storage medium based on Darkflow-DeepSort Download PDFInfo
- Publication number
- CN110516556A CN110516556A CN201910701678.5A CN201910701678A CN110516556A CN 110516556 A CN110516556 A CN 110516556A CN 201910701678 A CN201910701678 A CN 201910701678A CN 110516556 A CN110516556 A CN 110516556A
- Authority
- CN
- China
- Prior art keywords
- target
- darkflow
- deepsort
- appearance features
- frame
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 98
- 238000000034 method Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000004590 computer program Methods 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 15
- 238000001914 filtration Methods 0.000 description 13
- 230000000694 effects Effects 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005352 clarification Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000001973 epigenetic effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000005086 pumping Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035807 sensation Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
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
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- 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]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of multi-target tracking detection method, device and storage medium based on Darkflow-DeepSort, be related to intelligent Decision Technology field, method therein the following steps are included: S110, using YOLOv3 algorithm training obtain the target detection model based on Darkflow；S120, it will test image and input the trained target detection model based on Darkflow, obtain the appearance features of multiple targets；Wherein, the detection image is based on being decoded acquisition to monitor video；S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort；The target following model is obtained by the data set MOT16Challenge training of multi-target detection；S140, data correlation processing frame by frame is carried out to the monitor video using the Kalman filter of target following model, realizes the multi-target tracking in the monitor video.It is able to ascend multi-target tracking detection speed using foregoing invention, and completes multi-target tracking in the case where not losing accuracy in detection.
Description
Technical field
The present invention relates to intelligent Decision Technology fields, more specifically, are related to a kind of based on Darkflow-DeepSort's
Multi-target tracking detection method, device and storage medium.
Background technique
The method of sensation target tracking is widely used in the fields such as human-computer interaction, unmanned, is based on correlation filtering
The tracking of (Correlation Filter) and convolutional neural networks (CNN) has already taken up the more than half of target tracking domain
Rivers and mountains.
In existing multi-target tracking method SORT method (SIMPLE ONLINE AND REALTIME TRACKING,
Simple online and real-time tracking) the preferable effect that obtains.The feature of this method maximum is efficiently to realize target inspection
It surveys and goes filtering and Hungarian algorithm to be tracked using Kalman filtering.
And DeepSort is improvement on the basis of SORT target tracking, it is high performance using original DeepSort training
Faster-RCNN model carries out target detection, 45% ID switch is reduced relative to Sort algorithm, and combine depth
Appearance information is spent, has to the tracking effect of shelter target and greatly promotes；FP is increased, state-of-the-art online tracking effect has been reached
Fruit.But fps can reach 15fps or so when being tracked with the DeepSort of the method, but average fps is only capable of reaching
10 or so, and real-time tracing is only only stable in 8fps or so.
So needing a kind of promotion detection speed, and the multi-target tracking detection method of accuracy in detection is not lost.
Summary of the invention
To solve the above-mentioned problems, the object of the present invention is to provide a kind of multiple targets based on Darkflow-DeepSort
Tracking detection method, device and storage medium.
A kind of multi-target tracking detection method based on Darkflow-DeepSort is applied to electronic device, including following
Step:
S110, the target detection model based on Darkflow is obtained using the training of YOLOv3 algorithm；
S120, it will test image and input the trained target detection model based on Darkflow, obtain multiple targets
Appearance features；Wherein, the detection image is based on being decoded acquisition to monitor video；
S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort；It is described
Target following model is obtained by the data set MOT16Challenge training of multi-target detection；
S140, the monitor video is carried out at data correlation frame by frame using the Kalman filter of target following model
Reason, realizes the multi-target tracking in the monitor video.
Further, preferred method is that the target detection model based on Darkflow is Python model, described
Darknet network structure is converted by Cython and is obtained by Python model.
Further, preferred method is that the step S140 is specifically included:
S210, the motion match degree and appearance features matching degree for obtaining multiple target；Wherein, the motion match degree passes through
Calculating acquisition is carried out to the kinematic similarity for the multiple target that Kalman filter obtains；The appearance features matching degree is by by institute
The appearance features for stating multiple targets, which calculate, to be obtained；S220, motion match degree and appearance features matching degree using multiple target, pass through
To the data correlation processing of the monitor video frame by frame, the matching degree of target frame is obtained；S230, the final matching degree of selection reach pre-
If the target frame of match parameter is as target tracking result.
Further, preferred method is,
For the appearance features of multiple targets obtained in the step S120, screening frequency of occurrence is more than given threshold
Target assigns its priority by cascade matching to the target.
Further, preferred method is that the padding of convolutional layer is 1, Chi Hua in the network structure of the Darkflow
Layer is maximum pond.
A kind of electronic device, comprising: memory, processor and storage are in the memory and can be in the processor
The computer program of upper multi-target tracking detection method of the operation based on Darkflow-DeepSort, it is described to be based on Darkflow-
The computer program of the multi-target tracking detection method of DeepSort realizes following steps when being executed by the processor:
S110, the target detection model based on Darkflow is obtained using the training of YOLOv3 algorithm；
S120, it will test image and input the trained target detection model based on Darkflow, obtain multiple targets
Appearance features；Wherein, the detection image is based on being decoded acquisition to monitor video；
S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort；It is described
Target following model is obtained by the data set MOT16Challenge training of multi-target detection；
S140, the monitor video is carried out at data correlation frame by frame using the Kalman filter of target following model
Reason, realizes the multi-target tracking in the monitor video.
Further, preferred structure is that the target detection model based on Darkflow is Python model, described
Darknet network structure is converted by Cython and is obtained by Python model.
Further, preferred structure is that the step 140 includes:
S210, the motion match degree and appearance features matching degree for obtaining multiple target；Wherein, the motion match degree passes through
Calculating acquisition is carried out to the kinematic similarity for the multiple target that Kalman filter obtains；The appearance features matching degree is by by institute
The appearance features for stating multiple targets, which calculate, to be obtained；S220, motion match degree and appearance features matching degree using multiple target, pass through
To the data correlation processing of the monitor video frame by frame, the matching degree of target frame is obtained；S230, the final matching degree of selection reach pre-
If the target frame of match parameter is as target tracking result.
Further, preferred structure is that the padding of convolutional layer is 1, Chi Hua in the network structure of the Darkflow
Layer is maximum pond.
According to another aspect of the present invention, a kind of computer readable storage medium, the computer-readable storage medium are provided
Matter is stored with computer program, and the computer program includes the multi-target tracking detection journey based on Darkflow-DeepSort
Sequence, the multi-target tracking detection program based on Darkflow-DeepSort is above-mentioned when being executed by processor to be based on
Step in the multi-target tracking detection method of Darkflow-DeepSort.
It, can be with using above-mentioned multi-target tracking detection method, device and storage medium based on Darkflow-DeepSort
The effect of realization is as follows:
1, the data correlation of the present invention using single Kalman filtering for assuming method for tracing and frame by frame is realized in monitor video
Multi-target tracking, and YOLOv3 algorithm and Kalman filtering are combined together, both can with the tracking multiple target of high-accuracy,
Again can to avoid assume algorithms with measuring number and number of targets exponentially increases the huge drawback of bring calculation amount.
2, moving target is positioned in the movement destination image of acquisition using target detection iconic model, for continuously obtaining
The multiframe consecutive image obtained, namely for video, by positioning moving target in every frame image, to realize in video
The tracing detection of moving target behavior；Due to YOLOv3 algorithm process picture speed quickly, under the same conditions, be based on
YOLOv3 algorithm training objective detection model to image processing speed than existing convolutional neural networks algorithm training model (such as
1000 times faster than R-CNN, 100 times faster than Fast-RCNN) processing speed it is fast.
3, the transplanting of YOLOv3 algorithm is convenient, can realize under each operating system, to the configuration requirement phase of terminal hardware
To lower, the operation of target detection model can be easier realized in lightweight equipment.
4, the appearance features for extracting target to be tracked carry out arest neighbors matching, improve the target tracking under circumstance of occlusion
Effect, meanwhile, reduce the problem of Target id jumps.
5, it when using the target in method tracking video of the invention, in the video that original fps is 25, does not do and takes out at frame
In the case where reason, 15fps can achieve, it is optimal to can achieve 20 or more fps and lose when doing every three frames pumping frame processing
Track target；And 14fps or more can also can achieve for the tracking of real-time photography head, on the basis of guaranteeing accuracy, it will examine
Degree of testing the speed promotes 100 times.
6, for the application scenarios of real-time recorded broadcast, the present invention can realize the standard to moving target feature under identical precision
It determines position and quickly identification, improves the speed and precision identified in video field, reduce the delay and Caton of recording and broadcasting system.
To the accomplishment of the foregoing and related purposes, one or more aspects of the present invention includes the spy being particularly described below
Sign.Certain illustrative aspects of the invention is described in detail in the following description and the annexed drawings.However, these aspect instructions are only
It is that some of the various ways in the principles of the present invention can be used.In addition, the present invention is intended to include all such aspects with
And their equivalent.
Detailed description of the invention
By reference to the explanation below in conjunction with attached drawing, and with a more complete understanding of the present invention, of the invention is other
Purpose and result will be more clearly understood and understood.In the accompanying drawings:
Fig. 1 is the stream according to the multi-target tracking detection method based on Darkflow-DeepSort of the embodiment of the present invention
Cheng Tu；
The flow chart of the tracking of Fig. 2 target following model according to an embodiment of the present invention；
The conversion stream of Fig. 3 model structure according to an embodiment of the present invention that Darknet network structure is converted into Python
Journey schematic diagram；
Fig. 4 schematic network structure according to an embodiment of the present invention based on Darkflow；
The electronic device of Fig. 5 multi-target tracking detection according to an embodiment of the present invention based on Darkflow-DeepSort
Structural schematic diagram；
The process frame diagram of the tracking of Fig. 6 existing target tracking according to an embodiment of the present invention.
Identical label indicates similar or corresponding feature or function in all the appended drawings.
Specific embodiment
In the following description, for purposes of illustration, it in order to provide the comprehensive understanding to one or more embodiments, explains
Many details are stated.It may be evident, however, that these embodiments can also be realized without these specific details.
In other examples, one or more embodiments for ease of description, well known structure and equipment are shown in block form an.
The present invention provides a kind of multi-target tracking detection method based on Darkflow-DeepSort, electronic device and deposits
Storage media.Multi-target tracking detection method therein based on Darkflow-DeepSort, including target detection stage and target
Track phase；Be related to two models of Darkflow and DeepSort, Darkflow model therein be mainly used for training sample into
Tracking section such as Kalman filter confirmation track etc. is only used only in row pedestrian detection, DeepSort model.It is provided by the invention
Multi-target tracking detection method based on Darkflow-DeepSort is on the basis for extracting space characteristics using convolutional neural networks
On, using the characteristics of motion of Kalman filter learning objective, clarification of objective is merged, the position of target is carried out pre-
It calculates, the similarity that target is calculated in terms of binding time and space two carries out object matching, realizes the purpose of target tracking.
Fig. 1 shows the stream of the multi-target tracking detection method based on Darkflow-DeepSort of the embodiment of the present invention
Journey.
As shown in Figure 1, the multi-target tracking detection method based on Darkflow-DeepSort, includes the following steps:
S110, the target detection model based on Darkflow is obtained using the training of YOLOv3 algorithm；Wherein, Darkflow
Model is based on YOLOv3 algorithm, using the two-value cross entropy training gained of definition；And loss function when YOLOv3 algorithm
For two parts, first part is classification error, and it is all the two-value cross entropy of definition, then that second part, which is object space error,
Take the sum of squares of deviations of two class errors as total error function.
It should be noted that YOLOv3 (You only look once v3) is a kind of target based on Darknet-53
Detection algorithm, for other deep learning algorithms, promoted in performance it is maximum be exactly its detect speed faster, this
It is also the reason of we want in this way as the target detection in the present invention in multi-target tracking detection.Wherein, In
The target detection stage carries out target detection using the Darkflow model of YOLOv3 algorithm training, takes Darkflow network knot
Network frame of the structure as target detection；In the target tracking stage, target tracking is completed using the model of Python.
S120, it will test image and input the trained target detection model based on Darkflow, obtain multiple targets
Appearance features.
Wherein, the detection image is based on being decoded acquisition to monitor video；Illustratively be described as follows: to video into
The decoded ways customary of row is every frame decoding, for example, decoding interval frame number is arranged on the basis of 4 frames of extraction per second, if video fps
It is 24, then interval frame number is 6, goes out image to video real-time decoding using Opencv according to decoding interval frame number.
Appearance features i.e. location information and space characteristics；Further, the target detection based on Darkflow
Model is Python model, and Python model is namely based on the target detection model of Darkflow, in this module of target detection
It is down exactly depth characteristic describer；Appearance features are extracted using depth characteristic describer.Illustratively, using 8 parametersThe description of motion state is carried out, wherein (u, v) be the centre coordinate of bounding box,
R is length-width ratio, and h indicates height.Remaining four variable indicates the corresponding velocity information in image coordinate system.Wherein,
The appearance features of bounding box are the features of 128 dimensions obtained by a depth network.
S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort, it is described
Target following model is obtained by the data set MOT16Challenge training of multi-target detection.And target tracking model just utilizes
Remove the part of target detection in DeepSort model, that is, is matched including Kalman filter and subsequent cascaded.
Target following model is to determine the location information of target in the video frame, need apparent by corresponding target
Character description method comes out wherein metastable statistical nature or certain constant feature extractions, is obtained by filter
The response of object candidate area, as the standard for judging target position.
Target following model based on DeepSort is the data set MOT16Challenge in disclosed multi-target detection
What upper training obtained.Training set itself is the competition data for being exactly MOT16Challenge offer.
Further, the target following model foundation based on DeepSort is used on the basis of Kalman filter
Kalman filter goes to establish trace model, then realizes that target detection determines that appearance features are matched by Darkflow,
And location information input Kalman filter is tracked.
S140, the monitor video is carried out at data correlation frame by frame using the Kalman filter of target following model
Reason realizes the multi-target tracking in monitor video.Multi-target tracking is realized using single Kalman filtering for assuming method for tracing, both
Can with the tracking multiple target of high-accuracy, and can to avoid assume algorithms with measuring number and number of targets exponentially increases
The huge drawback of calculation amount.
Fig. 2 shows the processes of the working method of the target tracking model based on DeepSort of the embodiment of the present invention.
As shown in Fig. 2, the multi-target tracking detection method based on DeepSort, and the core concept of DeepSort is single vacation
If method for tracing, data correlation using recursive Kalman filtering and frame by frame realizes the tracing process of multiple target.It needs to illustrate
, DeepSort introduce pedestrian identify again data set (ReID data set, comprising 1261 people be more than 1,100,000 images,
The data set is suitble to do pedestrian's tracking) on off-line training deep learning model.The target based on DeepSort in the present invention
Tracing model just utilizes the part for removing target detection in DeepSort model.
In the target tracking stage, visual target tracking task be exactly the target sizes that give certain video sequence initial frame with
In the case where position, size and the position of the target of subsequent frame are predicted.And visual target tracking flow of task is divided according to frame,
As shown in Figure 6.Initialized target frame first generates numerous candidate frames in the next frame, extracts the feature of these candidate frames, then
It scores these candidate frames, target of the candidate frame of a highest scoring as prediction is finally found in these scorings, or
Multiple predicted values are merged and are more preferably predicted target.In the present invention, the prediction of target is to pass through Kalman filtering
Device is realized.
Wherein, a representative instance of Kalman filtering (Kalman) is limited from one group, comprising noise, to object
The observation sequence (may have deviation) of position predicts the coordinate and speed of the position of object.Many engineer applications (such as radar,
Computer vision) in can find its figure.Meanwhile Kalman filtering is also in control theory and control systems engineering (CSE)
An important topic.For example, people are interested to be their ability to tracking target for radar.But position, the speed of target
It spends, the measured value of acceleration often has noise at any time.Kalman filtering utilizes the multidate information of target, tries to remove
The influence of noise obtains the good estimation about target position.This estimation can be the estimation to current goal position
(filtering) is also possible to the estimation (prediction) for position in future, is also possible to the estimation to past position (interpolation or smooth).
The appearance features for the target that target detection model is obtained carry out arest neighbors matching by Kalman filter；It should be noted that
Arest neighbors matching is exactly to find a nearest feature according to the distance of feature to complete matching.The prediction of the position of this feature
It is to be completed by Kalman filter, then does arest neighbors matching with the position of the target detected in practice.
S210, the motion match degree and appearance features matching degree for obtaining multiple target；Wherein, the motion match degree passes through
Calculating acquisition is carried out to the kinematic similarity for the multiple target that Kalman filter obtains；The appearance features matching degree is by by institute
The appearance features for stating multiple targets, which calculate, to be obtained；S220, motion match degree and appearance features matching degree using multiple target, pass through
Data correlation processing frame by frame is carried out to monitor video, obtains the matching degree of target frame；S230, the final matching degree of selection reach pre-
If the target frame of match parameter is as target tracking result.That is, target and tracker are matched, update is paired into
Function is unsatisfactory for the tracking of condition with failed tracking, deletion is matched；Then technology is carried out to target and draws track, to complete
The tracking of target acts.
It should be noted that the tracking object of target frame can be people, the object that animal can also be moved with other can be.
When tracking object is people, target frame is properly termed as human body frame.
In the particular embodiment, multi-target tracking, the judgement point of matching degree are completed by the matching degree judgement of target frame
For two parts: IOU is matched and the matching of appearance features；Wherein, the matching of IOU is carried out between detecting twice to front and back
IOU matching；The matching of appearance features is that appearance features vector is extracted by a network, current to track target and potential
It will do it comparison with object.Calculate the front and back small minimum value of appearance features vector average distance twice.Wherein, of appearance features
With degree=1- normalization average distance minimum value.
And final matching degree is equal to the average value of IOU matching value and appearance features matching value, that is to say, that final matching degree
Equal to (IOU matching value+appearance features matching value)/2.
Preset matching parameter is that final matching degree is greater than 0.5, and IOU matching value is greater than 0.5；If having reached preset matching
Parameter then illustrates successful match, for tracking；Otherwise, it is determined that unsuccessful to match.
For updating successful matching and matching failed tracking, the case where being unsatisfactory for the tracking of condition specifics is deleted
Illustrative explanation, the prediction of target state is carried out using a standard Kalman filter, wherein Kalman filter
For based on constant rate pattern (it is constant just to refer to that speed is defaulted as, that is, the model of acceleration is not present) and Systems with Linear Observation mould
Type.
The result of the prediction of Kalman filter is (u, v, r, h), tracks target to each, record is examined from its last time
Survey the frame number a after result is matched with tracking resultk·Once the testing result of a target is correctly associated with it with tracking result
Afterwards, just by parameter ak·It is set as 0.Wherein, the logger or array that record is just comparable to an outside go to record
Each frame, the tracking data of each target, Kalman filtering are only then to do pre- by the position of each target of input
It surveys.
It should be noted that the predicted value and actually detected value to Kalman filter compare, if observation and pre-
Measured value has big difference, then prediction cannot represent observation.
That is, Amax is a upper limit, ak·It is then kalman filter prediction and the unmatched frame number of observation
Value, if ak·It has been more than Amax, then has illustrated that Kalman filter tracking effect is bad.Then think the tracking to the target
Journey has terminated, and does not just continue to track.That is, tracking process terminates to refer to that We conducted track still to a target
Subsequent Kalman filter can not accurately be predicted after new position, it is believed that tracking finishes.
Wherein, to fresh target occur judgement be then, if some target in certain testing result always can not with
It is closed through existing tracker (being detected before already present tracker, tracking the tracker of target now)
Connection, then then thinking to be likely to occur fresh target.
If (new tracker, can aiming at the target of new appearance for potential new tracker in continuous 3 frame
Continuous three frames interaction prediction result and testing result, then it is assumed that be new tracker.) can to the prediction result of target position
It is correctly associated with testing result, then then confirmation is that new moving target occurred.
If the requirement cannot be reached, then it is assumed that be " false-alarm " occurred, need to delete the moving target；That is,
Refer to when deleting the moving target, cannot complete to match in its continuous three frame of the target that the detection model new for one detects,
It is tracking target (being probably derived from detection error) that we, which are considered as this target not, deletes this target.
Whether the appearance of one new target can be matched first with already existing tracker, see before belonging to and tracking
Target need to create new tracker if it is not, then being considered to be likely to occur new target.
In one particular embodiment of the present invention, for the apparent spy of multiple targets obtained in the step S120
Sign, screening frequency of occurrence are more than the target of given threshold, assign its priority by cascade matching to the target.Wherein, out
The given threshold of occurrence number is usually set to 3 times.
Further, matched final stage is being cascaded, in order to alleviate because caused by epigenetic mutation or partial occlusion
Large change, can carry out the matching based on IOU to the track that do not match of unconfirmed and age=1.
Assigning priority to the target frequently occurred by cascade matching is the shape blocked for a long time for a target
State setting.Wherein, after a target is blocked for a long time, the uncertainty of Kalman prediction will increase
Add, the observability in state space just will be greatly reduced.
If two trackers compete the matching power of the same testing result at this time, that longer rail of time is often blocked
The mahalanobis distance of mark is smaller, so that testing result, which is more likely to and blocks that longer track of time, to be associated, it is this undesirable
Effect often cracking tracing duration.
That is, it is assumed that original covariance matrix is a normal distribution, then continuous prediction is not updated and will be led
Cause the variance of this normal distribution increasing, then the point remote from mean value Euclidean distance may with before distribution in from it is closer
Point obtain same mahalanobis distance value.Come so cascade matching (Matching Cascade) is employed herein to more
The target frequently occurred assigns priority.
It should be noted that the matched meaning of cascade is exactly that various matching ways are combined to (such as IOU matching or feature
With), it is matched by cascade mode (i.e. a matching way connects a matching way)；Alternatively, being further first added
Criterion is selected, corresponding matching is then carried out.
In a specific embodiment, using second of cascade system, that is to say, that be first added and select criterion, then
Carry out corresponding matching movement.Therefore, it first joined a time point sequence, preferentially chosen from the high target of the frequency of occurrences
Choosing, subsequently into matching mechanisms, so that the target being blocked for a long time is more difficult by priority match, i.e., to more frequently occurring
Target assign priority.
Specific algorithm is referring to paper SIMPLE ONLINE AND REALTIME TRACKING WITH ADEEP
ASSOCIATION METRIC；Nicolai Wojke, Alex Bewley, Dietrich Paulus, University of
Koblenz-Landau, Queensland University of Technology, details are not described herein.
Fig. 3 shows the conversion of the model structure that Darknet network structure is converted into Python of the embodiment of the present invention
Process；
As shown in figure 3, Darknet network structure to be converted into the flow path switch of the model structure of Python, including as follows
Step:
That is Darknet is translated as by stream used in Tensorflow by Darkflow；Darkflow will
Darknet is translated as the process of Tensorflow.
By Cython, former base is facilitated confession in the model structure that the Darknet network structure of C is converted into Python by us
DeepSort is used.It is used it is also possible to generate the pb model structure that Tensorflow is used for other algorithms.
Fig. 4 shows the network structure based on Darkflow of the embodiment of the present invention；
As shown in figure 4, the network structure of Darkflow is as follows:
The padding of all convolutional layers is 1 in Darkflow network structure, and pond layer is maximum pond.Other
Parameter such as step-length, convolution kernel size, the number of filter, as shown in the figure.
It is most initially convolution kernel is the convolutional layer that (3*3) filter number is 32；It is 2, Chi Hua great followed by a step-length
The small maximum pond for being 2；It is a convolution kernel is later the convolutional layer that (3*3) filter number is 64, and then a step-length is
2, the maximum pond that size is 2.
Network structure later is similar, is all first to carry out the convolution that a convolution kernel is N for (3*3) filter number
Layer, wherein N is two times of filter number of last big convolutional coding structure.Then carrying out a convolution kernel is (1*1)
Filter number is the convolution of N/2, then carrying out a convolution kernel is the convolutional layer that (3*3) filter number is N, finally carries out one
Secondary maximum pond.Form a big convolutional coding structure.The convolutional coding structure carries out 4 times altogether, removes pond layer in last time,
Connect two corresponding convolutional layers.
Corresponding with the above-mentioned multi-target tracking detection method based on Darkflow-DeepSort, the invention also includes bases
In the multi-target tracking detection system of Darkflow-DeepSort, including:
Including target detection model training unit, for obtaining the target based on Darkflow using the training of YOLOv3 algorithm
Detection model；
Appearance features determination unit inputs the trained target detection mould based on Darkflow for will test image
Type obtains the appearance features of multiple targets；Detection image therein is based on being decoded acquisition to monitor video；
Target tracking model training unit obtains target by the data set MOT16Challenge training of multi-target detection
The appearance features of multiple targets are inputted the trained target following model based on DeepSort by trace model.
Target Acquisition unit carries out number frame by frame to the monitor video using the Kalman filter of target following model
According to association process, the multi-target tracking in monitor video is realized.
Wherein, it is based in the specific implementation function and embodiment of multi-target detection unit and multi-target tracking unit
The corresponding step of the multi-target tracking detection method of Darkflow-DeepSort corresponds, and the present embodiment is not described in detail one by one.
Fig. 5 is the schematic diagram for the electronic device logical construction that one embodiment of the invention provides.
As shown in figure 5, the electronic device 50 of the embodiment is including processor 51, memory 52 and is stored in memory 52
In and the computer program 53 that can be run on processor 51.Processor 51 realizes base in embodiment when executing computer program 53
In each step of the multi-target tracking detection method of Darkflow-DeepSort, such as step S110 shown in FIG. 1 is extremely
S140.Alternatively, processor 51 realizes above-mentioned each dress when executing the multi-target tracking detection method based on Darkflow-DeepSort
Set the function of each module/unit in embodiment.
Illustratively, computer program 53 can be divided into one or more module/units, one or more mould
Block/unit is stored in memory 52, and is executed by processor 51, to complete the present invention.One or more module/units can
To be the series of computation machine program instruction section that can complete specific function, the instruction segment is for describing computer program 53 in electricity
Implementation procedure in sub-device 50.For example, computer program 53 can be divided into multi-target detection unit and multi-target tracking
Unit, function have a detailed description in embodiment, will not repeat them here.
Electronic device 50 can be desktop PC, notebook, palm PC and cloud server etc. and calculate equipment.Electricity
Sub-device 50 may include, but be not limited only to, processor 51, memory 52.It will be understood by those skilled in the art that Fig. 5 is only
The example of electronic device 50 does not constitute the restriction to electronic device 50, may include components more more or fewer than diagram, or
Person combines certain components or different components, such as electronic device can also be set including input-output equipment, network insertion
Standby, bus etc..
Alleged processor 51 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor is also possible to any conventional processor
Deng.
Memory 52 can be the internal storage unit of electronic device 50, such as the hard disk or memory of electronic device 50.It deposits
Reservoir 52 is also possible to the plug-in type hard disk being equipped on the External memory equipment of electronic device 50, such as electronic device 50, intelligence
Storage card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card (Flash Card)
Deng.Further, memory 52 can also both including electronic device 50 internal storage unit and also including External memory equipment.It deposits
Reservoir 52 is for storing other programs and data needed for computer program and electronic equipment.Memory 52 can be also used for temporarily
When store the data that has exported or will export.
The present embodiment provides a computer readable storage medium, computer journey is stored on the computer readable storage medium
Sequence realizes the multi-target tracking inspection in embodiment based on Darkflow-DeepSort when the computer program is executed by processor
Survey method, to avoid repeating, which is not described herein again.Alternatively, the computer program realizes above-mentioned be based on when being executed by processor
The function of each module/unit in the multi-target tracking detection system of Darkflow-DeepSort, to avoid repeating, here no longer
It repeats.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of device are divided into different functional unit or module, to complete above description
All or part of function.Each functional unit in embodiment, module can integrate in one processing unit, be also possible to
Each unit physically exists alone, and can also be integrated in one unit with two or more units, above-mentioned integrated unit
Both it can take the form of hardware realization, can also realize in the form of software functional units.In addition, each functional unit, mould
The specific name of block is also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.It is single in above system
Member, the specific work process of module, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In embodiment provided by the present invention, it should be understood that disclosed device and method can pass through others
Mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of the module or unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be with
In conjunction with or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or discussed
Mutual coupling or direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING of device or unit or
Communication connection can be electrical property, mechanical or other forms.
Unit may or may not be physically separated as illustrated by the separation member, shown as a unit
Component may or may not be physical unit, it can and it is in one place, or may be distributed over multiple networks
On unit.It can some or all of the units may be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated module/unit be realized in the form of SFU software functional unit and as independent product sale or
In use, can store in a computer readable storage medium.Based on this understanding, the present invention realizes above-mentioned implementation
All or part of the process in example method, can also instruct relevant hardware to complete, the meter by computer program
Calculation machine program can be stored in a computer readable storage medium, the computer program when being executed by processor, it can be achieved that on
The step of stating each embodiment of the method.Wherein, the computer program includes computer program code, the computer program generation
Code can be source code form, object identification code form, executable file or certain intermediate forms etc..The computer-readable medium
It may include: any entity or device, recording medium, USB flash disk, mobile hard disk, magnetic that can carry the computer program code
Dish, CD, computer storage, read-only memory (ROM, Read-Only Memory), random access memory (RAM,
Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..It should be noted that described
The content that computer-readable medium includes can carry out increasing appropriate according to the requirement made laws in jurisdiction with patent practice
Subtract, such as in certain jurisdictions, according to legislation and patent practice, computer-readable medium do not include be electric carrier signal and
Telecommunication signal.
It is described in an illustrative manner above with reference to Fig. 1-Fig. 6 according to the present invention based on the more of Darkflow-DeepSort
Target tracking detection method, electronic device and storage medium.It will be understood by those skilled in the art, however, that being sent out for above-mentioned
Bright proposed multi-target tracking detection method, device and the storage medium based on Darkflow-DeepSort, can also be not
Various improvement are made on the basis of disengaging the content of present invention.Therefore, protection scope of the present invention should be by the attached claims
The content of book determines.
Claims (10)
1. a kind of multi-target tracking detection method based on Darkflow-DeepSort, is applied to electronic device, feature exists
In, comprising the following steps:
S110, the target detection model based on Darkflow is obtained using the training of YOLOv3 algorithm；
S120, it will test image and input the trained target detection model based on Darkflow, obtain the apparent of multiple targets
Feature；Wherein, the detection image is based on being decoded acquisition to monitor video；
S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort；The target
Trace model is obtained by the data set MOT16Challenge training of multi-target detection；
S140, data correlation processing frame by frame is carried out to the monitor video using the Kalman filter of target following model,
Realize the multi-target tracking in the monitor video.
2. the multi-target tracking detection method according to claim 1 based on Darkflow-DeepSort, feature exist
In,
The target detection model based on Darkflow is Python model, and the Python model will by Cython
The conversion of Darknet network structure obtains.
3. the multi-target tracking detection method according to claim 1 based on Darkflow-DeepSort, feature exist
In the step 140 includes:
S210, the motion match degree and appearance features matching degree for obtaining multiple target；Wherein, the motion match degree passes through to card
The kinematic similarity for the multiple target that Thalmann filter obtains carries out calculating acquisition；The appearance features matching degree passes through will be described more
The appearance features of a target, which calculate, to be obtained；
S220, motion match degree and appearance features matching degree using multiple target, pass through the data to the monitor video frame by frame
Association process, obtained IOU matching value and appearance features matching value pass through IOU matching value and appearance features matching value meter
Calculate the final matching degree of target frame；
S230, the final matching degree of selection reach the target frame of preset matching parameter as target tracking result.
4. the multi-target tracking detection method according to claim 1 based on Darkflow-DeepSort, feature exist
In, for the appearance features of multiple targets obtained in the step S120, the target that frequency of occurrence is more than given threshold is screened,
Its priority is assigned by cascade matching to the target.
5. the multi-target tracking detection method according to claim 2 based on Darkflow-DeepSort, feature exist
In,
The padding of convolutional layer is 1 in the network structure of the Darkflow, and pond layer is maximum pond.
6. a kind of electronic device, characterized by comprising: memory, processor and storage are in the memory and can be in institute
The computer program that the multi-target tracking detection method based on Darkflow-DeepSort is run on processor is stated, it is described to be based on
Following step is realized when the computer program of the multi-target tracking detection method of Darkflow-DeepSort is executed by the processor
It is rapid:
S110, the target detection model based on Darkflow is obtained using the training of YOLOv3 algorithm；
S120, it will test image and input the trained target detection model based on Darkflow, obtain the apparent of multiple targets
Feature；Wherein, the detection image is based on being decoded acquisition to monitor video；
S130, the appearance features of multiple targets are inputted into the trained target following model based on DeepSort；The target
Trace model is obtained by the data set MOT16Challenge training of multi-target detection；
S140, data correlation processing frame by frame is carried out to the monitor video using the Kalman filter of target following model,
Realize the multi-target tracking in the monitor video.
7. electronic device according to claim 6, which is characterized in that the target detection model based on Darkflow is
Darknet network structure is converted by Cython and is obtained by Python model, the Python model.
8. electronic device according to claim 6, which is characterized in that the step 140 includes:
S210, the motion match degree and appearance features matching degree for obtaining multiple target；Wherein, the motion match degree passes through to card
The kinematic similarity for the multiple target that Thalmann filter obtains carries out calculating acquisition；The appearance features matching degree passes through will be described more
The appearance features of a target, which calculate, to be obtained；
S220, motion match degree and appearance features matching degree using multiple target, pass through the data to the monitor video frame by frame
Association process, obtained IOU matching value and appearance features matching value pass through IOU matching value and appearance features matching value meter
Calculate the final matching degree of target frame；
S230, the final matching degree of selection reach the target frame of preset matching parameter as target tracking result.
9. electronic device according to claim 7, which is characterized in that convolutional layer in the network structure of the Darkflow
Padding is 1, and pond layer is maximum pond.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, the calculating
Machine program includes the multi-target tracking detection program based on Darkflow-DeepSort, described to be based on Darkflow-DeepSort
Multi-target tracking detection program realize when being executed by processor and be based on as according to any one of claims 1 to 5
Step in the multi-target tracking detection method of Darkflow-DeepSort.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910701678.5A CN110516556B (en) | 2019-07-31 | 2019-07-31 | Multi-target tracking detection method and device based on Darkflow-deep Sort and storage medium |
| PCT/CN2019/117801 WO2021017291A1 (en) | 2019-07-31 | 2019-11-13 | Darkflow-deepsort-based multi-target tracking detection method, device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910701678.5A CN110516556B (en) | 2019-07-31 | 2019-07-31 | Multi-target tracking detection method and device based on Darkflow-deep Sort and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110516556A true CN110516556A (en) | 2019-11-29 |
| CN110516556B CN110516556B (en) | 2023-10-31 |
Family
ID=68624348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910701678.5A Active CN110516556B (en) | 2019-07-31 | 2019-07-31 | Multi-target tracking detection method and device based on Darkflow-deep Sort and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110516556B (en) |
| WO (1) | WO2021017291A1 (en) |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178208A (en) * | 2019-12-20 | 2020-05-19 | 华瑞新智科技（北京）有限公司 | Pedestrian detection method, device and medium based on deep learning |
| CN111222574A (en) * | 2020-01-07 | 2020-06-02 | 西北工业大学 | Ship and civil ship target detection and classification method based on multi-model decision-level fusion |
| CN111259868A (en) * | 2020-03-10 | 2020-06-09 | 北京以萨技术股份有限公司 | Convolutional neural network-based method, system and medium for detecting vehicles in reverse driving |
| CN111260628A (en) * | 2020-01-15 | 2020-06-09 | 北京林业大学 | Large nursery stock number counting method based on video image and electronic equipment |
| CN111340764A (en) * | 2020-02-20 | 2020-06-26 | 江苏东印智慧工程技术研究院有限公司 | Automatic inhaul cable apparent disease counting method based on deep sort |
| CN111382704A (en) * | 2020-03-10 | 2020-07-07 | 北京以萨技术股份有限公司 | Vehicle line-pressing violation judgment method and device based on deep learning and storage medium |
| CN111401285A (en) * | 2020-03-23 | 2020-07-10 | 北京迈格威科技有限公司 | Target tracking method and device and electronic equipment |
| CN111428644A (en) * | 2020-03-25 | 2020-07-17 | 北京以萨技术股份有限公司 | Zebra crossing region monitoring method, system and medium based on deep neural network |
| CN111428642A (en) * | 2020-03-24 | 2020-07-17 | 厦门市美亚柏科信息股份有限公司 | Multi-target tracking algorithm, electronic device and computer readable storage medium |
| CN111460968A (en) * | 2020-03-27 | 2020-07-28 | 上海大学 | Video-based unmanned aerial vehicle identification and tracking method and device |
| CN111708380A (en) * | 2020-06-29 | 2020-09-25 | 北京御航智能科技有限公司 | Detection method, platform, unmanned aerial vehicle and system for appearance defects of wind turbine generator |
| CN111723664A (en) * | 2020-05-19 | 2020-09-29 | 烟台市广智微芯智能科技有限责任公司 | Pedestrian counting method and system for open type area |
| CN112036271A (en) * | 2020-08-18 | 2020-12-04 | 汇纳科技股份有限公司 | Pedestrian re-identification method, system, medium and terminal based on Kalman filtering |
| CN112200021A (en) * | 2020-09-22 | 2021-01-08 | 燕山大学 | Target crowd tracking and monitoring method based on limited range scene |
| CN112241974A (en) * | 2020-05-29 | 2021-01-19 | 北京新能源汽车技术创新中心有限公司 | Traffic accident detection method, processing method, system and storage medium |
| CN112329521A (en) * | 2020-09-24 | 2021-02-05 | 上海品览数据科技有限公司 | Multi-target tracking video shop-patrol method based on deep learning |
| CN112668432A (en) * | 2020-12-22 | 2021-04-16 | 上海幻维数码创意科技股份有限公司 | Human body detection tracking method in ground interactive projection system based on YoloV5 and Deepsort |
| CN112767711A (en) * | 2021-01-27 | 2021-05-07 | 湖南优美科技发展有限公司 | Multi-class multi-scale multi-target snapshot method and system |
| CN112785625A (en) * | 2021-01-20 | 2021-05-11 | 北京百度网讯科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN112883871A (en) * | 2021-02-19 | 2021-06-01 | 北京三快在线科技有限公司 | Model training and unmanned vehicle motion strategy determining method and device |
| CN112926649A (en) * | 2021-02-24 | 2021-06-08 | 北京优创新港科技股份有限公司 | Method and device for recognizing repeated weighing behaviors of cigarette frame |
| CN112950671A (en) * | 2020-08-06 | 2021-06-11 | 郑锴 | Real-time high-precision parameter measurement method for moving target by unmanned aerial vehicle |
| CN112985439A (en) * | 2021-02-08 | 2021-06-18 | 青岛大学 | Pedestrian jam state prediction method based on YOLOv3 and Kalman filtering |
| CN113076899A (en) * | 2021-04-12 | 2021-07-06 | 华南理工大学 | High-voltage transmission line foreign matter detection method based on target tracking algorithm |
| CN113139442A (en) * | 2021-04-07 | 2021-07-20 | 青岛以萨数据技术有限公司 | Image tracking method and device, storage medium and electronic equipment |
| CN113158995A (en) * | 2021-05-21 | 2021-07-23 | 西安建筑科技大学 | Multi-target tracking detection method, system, equipment and storage medium |
| CN113256690A (en) * | 2021-06-16 | 2021-08-13 | 中国人民解放军国防科技大学 | Pedestrian multi-target tracking method based on video monitoring |
| CN113591577A (en) * | 2021-06-30 | 2021-11-02 | 安徽省国维通信工程有限责任公司 | Artificial intelligence video image detecting system in communication engineering |
| CN113701758A (en) * | 2021-08-23 | 2021-11-26 | 中国北方工业有限公司 | Multi-target data association method and system based on biological search algorithm |
| CN113781521A (en) * | 2021-07-12 | 2021-12-10 | 山东建筑大学 | Improved YOLO-Deepsort-based bionic robot fish detection and tracking method |
| CN114155275A (en) * | 2021-11-17 | 2022-03-08 | 深圳职业技术学院 | IOU-Tracker-based fish tracking method and device |
| CN114524339A (en) * | 2022-01-06 | 2022-05-24 | 广东博智林机器人有限公司 | Elevator car safe operation detection method, device, equipment and storage medium |
| CN115049924A (en) * | 2022-06-06 | 2022-09-13 | 四川大学 | Building earthquake damage assessment method based on non-structural component damage identification under video monitoring |
| CN115082526A (en) * | 2022-07-26 | 2022-09-20 | 复亚智能科技(太仓)有限公司 | Target tracking method and device |
| CN116069801A (en) * | 2023-03-06 | 2023-05-05 | 山东华夏高科信息股份有限公司 | Traffic video structured data generation method, device and medium |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113341391B (en) * | 2021-06-01 | 2022-05-10 | 电子科技大学 | Radar target multi-frame joint detection method in unknown environment based on deep learning |
| CN113343836A (en) * | 2021-06-02 | 2021-09-03 | 禾麦科技开发（深圳）有限公司 | Floor elevator waiting crowd detection system and method based on convolutional neural network |
| CN113420679B (en) * | 2021-06-26 | 2024-04-26 | 南京搜文信息技术有限公司 | Artificial intelligence cross-camera multi-target tracking system and tracking method |
| CN113470078A (en) * | 2021-07-15 | 2021-10-01 | 浙江大华技术股份有限公司 | Target tracking method, device and system |
| CN113822153A (en) * | 2021-08-11 | 2021-12-21 | 桂林电子科技大学 | Unmanned aerial vehicle tracking method based on improved DeepSORT algorithm |
| CN113962282B (en) * | 2021-08-19 | 2024-04-16 | 大连海事大学 | Ship cabin fire real-time detection system and method based on improved yolov5l+deep |
| CN113674321B (en) * | 2021-08-25 | 2024-05-17 | 燕山大学 | Cloud-based method for multi-target tracking under monitoring video |
| CN114022812B (en) * | 2021-11-01 | 2024-05-10 | 大连理工大学 | DeepSort water surface floater multi-target tracking method based on lightweight SSD |
| CN114240996B (en) * | 2021-11-16 | 2024-05-07 | 灵译脑科技(上海)有限公司 | Multi-target tracking method based on target motion prediction |
| CN114820699B (en) * | 2022-03-29 | 2023-07-18 | 小米汽车科技有限公司 | Multi-target tracking method, device, equipment and medium |
| CN116777950B (en) * | 2023-04-19 | 2024-05-03 | 长沙理工大学 | Multi-target visual tracking method, device, equipment and medium based on camera parameters |
| CN117739994B (en) * | 2024-02-20 | 2024-04-30 | 广东电网有限责任公司阳江供电局 | Visual robot underwater target identification tracking method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019041519A1 (en) * | 2017-08-29 | 2019-03-07 | 平安科技（深圳）有限公司 | Target tracking device and method, and computer-readable storage medium |
| CN109816690A (en) * | 2018-12-25 | 2019-05-28 | 北京飞搜科技有限公司 | Multi-target tracking method and system based on depth characteristic |
| CN109871763A (en) * | 2019-01-16 | 2019-06-11 | 清华大学 | A kind of specific objective tracking based on YOLO |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11144761B2 (en) * | 2016-04-04 | 2021-10-12 | Xerox Corporation | Deep data association for online multi-class multi-object tracking |
| CN110047095B (en) * | 2019-03-06 | 2023-07-21 | 平安科技（深圳）有限公司 | Tracking method and device based on target detection and terminal equipment |
| CN109977818A (en) * | 2019-03-14 | 2019-07-05 | 上海极链网络科技有限公司 | A kind of action identification method and system based on space characteristics and multi-target detection |
-
2019
- 2019-07-31 CN CN201910701678.5A patent/CN110516556B/en active Active
- 2019-11-13 WO PCT/CN2019/117801 patent/WO2021017291A1/en active Application Filing
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019041519A1 (en) * | 2017-08-29 | 2019-03-07 | 平安科技（深圳）有限公司 | Target tracking device and method, and computer-readable storage medium |
| CN109816690A (en) * | 2018-12-25 | 2019-05-28 | 北京飞搜科技有限公司 | Multi-target tracking method and system based on depth characteristic |
| CN109871763A (en) * | 2019-01-16 | 2019-06-11 | 清华大学 | A kind of specific objective tracking based on YOLO |
Non-Patent Citations (1)
| Title |
|---|
| WOJKE, NICOLAI等: ""Simple Online and Realtime Tracking with a Deep Association Metrlc"", HTTPS://ARXIV.ORG/ABS/1703.07402V1, pages 1 - 5 * |
Cited By (52)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178208B (en) * | 2019-12-20 | 2023-08-15 | 华瑞新智科技（北京）有限公司 | Pedestrian detection method, device and medium based on deep learning |
| CN111178208A (en) * | 2019-12-20 | 2020-05-19 | 华瑞新智科技（北京）有限公司 | Pedestrian detection method, device and medium based on deep learning |
| CN111222574B (en) * | 2020-01-07 | 2022-04-05 | 西北工业大学 | Ship and civil ship target detection and classification method based on multi-model decision-level fusion |
| CN111222574A (en) * | 2020-01-07 | 2020-06-02 | 西北工业大学 | Ship and civil ship target detection and classification method based on multi-model decision-level fusion |
| CN111260628A (en) * | 2020-01-15 | 2020-06-09 | 北京林业大学 | Large nursery stock number counting method based on video image and electronic equipment |
| CN111340764A (en) * | 2020-02-20 | 2020-06-26 | 江苏东印智慧工程技术研究院有限公司 | Automatic inhaul cable apparent disease counting method based on deep sort |
| CN111259868B (en) * | 2020-03-10 | 2023-12-12 | 以萨技术股份有限公司 | Reverse vehicle detection method, system and medium based on convolutional neural network |
| CN111382704A (en) * | 2020-03-10 | 2020-07-07 | 北京以萨技术股份有限公司 | Vehicle line-pressing violation judgment method and device based on deep learning and storage medium |
| CN111382704B (en) * | 2020-03-10 | 2023-12-15 | 以萨技术股份有限公司 | Vehicle line pressing violation judging method and device based on deep learning and storage medium |
| CN111259868A (en) * | 2020-03-10 | 2020-06-09 | 北京以萨技术股份有限公司 | Convolutional neural network-based method, system and medium for detecting vehicles in reverse driving |
| CN111401285A (en) * | 2020-03-23 | 2020-07-10 | 北京迈格威科技有限公司 | Target tracking method and device and electronic equipment |
| CN111401285B (en) * | 2020-03-23 | 2024-02-23 | 北京迈格威科技有限公司 | Target tracking method and device and electronic equipment |
| CN111428642A (en) * | 2020-03-24 | 2020-07-17 | 厦门市美亚柏科信息股份有限公司 | Multi-target tracking algorithm, electronic device and computer readable storage medium |
| CN111428644A (en) * | 2020-03-25 | 2020-07-17 | 北京以萨技术股份有限公司 | Zebra crossing region monitoring method, system and medium based on deep neural network |
| CN111460968A (en) * | 2020-03-27 | 2020-07-28 | 上海大学 | Video-based unmanned aerial vehicle identification and tracking method and device |
| CN111460968B (en) * | 2020-03-27 | 2024-02-06 | 上海大学 | Unmanned aerial vehicle identification and tracking method and device based on video |
| CN111723664A (en) * | 2020-05-19 | 2020-09-29 | 烟台市广智微芯智能科技有限责任公司 | Pedestrian counting method and system for open type area |
| CN112241974B (en) * | 2020-05-29 | 2024-05-10 | 北京国家新能源汽车技术创新中心有限公司 | Traffic accident detection method, processing method, system and storage medium |
| CN112241974A (en) * | 2020-05-29 | 2021-01-19 | 北京新能源汽车技术创新中心有限公司 | Traffic accident detection method, processing method, system and storage medium |
| CN111708380B (en) * | 2020-06-29 | 2023-11-10 | 北京御航智能科技有限公司 | Wind turbine generator appearance defect detection method, platform, unmanned aerial vehicle and system |
| CN111708380A (en) * | 2020-06-29 | 2020-09-25 | 北京御航智能科技有限公司 | Detection method, platform, unmanned aerial vehicle and system for appearance defects of wind turbine generator |
| CN112950671B (en) * | 2020-08-06 | 2024-02-13 | 中国人民解放军32146部队 | Real-time high-precision parameter measurement method for moving target by unmanned aerial vehicle |
| CN112950671A (en) * | 2020-08-06 | 2021-06-11 | 郑锴 | Real-time high-precision parameter measurement method for moving target by unmanned aerial vehicle |
| CN112036271B (en) * | 2020-08-18 | 2023-10-10 | 汇纳科技股份有限公司 | Pedestrian re-identification method, system, medium and terminal based on Kalman filtering |
| CN112036271A (en) * | 2020-08-18 | 2020-12-04 | 汇纳科技股份有限公司 | Pedestrian re-identification method, system, medium and terminal based on Kalman filtering |
| CN112200021A (en) * | 2020-09-22 | 2021-01-08 | 燕山大学 | Target crowd tracking and monitoring method based on limited range scene |
| CN112329521A (en) * | 2020-09-24 | 2021-02-05 | 上海品览数据科技有限公司 | Multi-target tracking video shop-patrol method based on deep learning |
| CN112668432A (en) * | 2020-12-22 | 2021-04-16 | 上海幻维数码创意科技股份有限公司 | Human body detection tracking method in ground interactive projection system based on YoloV5 and Deepsort |
| CN112785625A (en) * | 2021-01-20 | 2021-05-11 | 北京百度网讯科技有限公司 | Target tracking method and device, electronic equipment and storage medium |
| CN112785625B (en) * | 2021-01-20 | 2023-09-22 | 北京百度网讯科技有限公司 | Target tracking method, device, electronic equipment and storage medium |
| CN112767711A (en) * | 2021-01-27 | 2021-05-07 | 湖南优美科技发展有限公司 | Multi-class multi-scale multi-target snapshot method and system |
| CN112767711B (en) * | 2021-01-27 | 2022-05-27 | 湖南优美科技发展有限公司 | Multi-class multi-scale multi-target snapshot method and system |
| CN112985439A (en) * | 2021-02-08 | 2021-06-18 | 青岛大学 | Pedestrian jam state prediction method based on YOLOv3 and Kalman filtering |
| CN112985439B (en) * | 2021-02-08 | 2023-10-17 | 青岛大学 | Pedestrian blocking state prediction method based on YOLOv3 and Kalman filtering |
| CN112883871A (en) * | 2021-02-19 | 2021-06-01 | 北京三快在线科技有限公司 | Model training and unmanned vehicle motion strategy determining method and device |
| CN112926649A (en) * | 2021-02-24 | 2021-06-08 | 北京优创新港科技股份有限公司 | Method and device for recognizing repeated weighing behaviors of cigarette frame |
| CN113139442A (en) * | 2021-04-07 | 2021-07-20 | 青岛以萨数据技术有限公司 | Image tracking method and device, storage medium and electronic equipment |
| CN113076899A (en) * | 2021-04-12 | 2021-07-06 | 华南理工大学 | High-voltage transmission line foreign matter detection method based on target tracking algorithm |
| CN113158995A (en) * | 2021-05-21 | 2021-07-23 | 西安建筑科技大学 | Multi-target tracking detection method, system, equipment and storage medium |
| CN113256690A (en) * | 2021-06-16 | 2021-08-13 | 中国人民解放军国防科技大学 | Pedestrian multi-target tracking method based on video monitoring |
| CN113256690B (en) * | 2021-06-16 | 2021-09-17 | 中国人民解放军国防科技大学 | Pedestrian multi-target tracking method based on video monitoring |
| CN113591577A (en) * | 2021-06-30 | 2021-11-02 | 安徽省国维通信工程有限责任公司 | Artificial intelligence video image detecting system in communication engineering |
| CN113781521B (en) * | 2021-07-12 | 2023-08-08 | 山东建筑大学 | Bionic robot fish detection tracking method based on improved YOLO-deep |
| CN113781521A (en) * | 2021-07-12 | 2021-12-10 | 山东建筑大学 | Improved YOLO-Deepsort-based bionic robot fish detection and tracking method |
| CN113701758A (en) * | 2021-08-23 | 2021-11-26 | 中国北方工业有限公司 | Multi-target data association method and system based on biological search algorithm |
| CN114155275A (en) * | 2021-11-17 | 2022-03-08 | 深圳职业技术学院 | IOU-Tracker-based fish tracking method and device |
| CN114524339A (en) * | 2022-01-06 | 2022-05-24 | 广东博智林机器人有限公司 | Elevator car safe operation detection method, device, equipment and storage medium |
| CN114524339B (en) * | 2022-01-06 | 2024-02-09 | 广东博智林机器人有限公司 | Method, device, equipment and storage medium for detecting safe operation of elevator car |
| CN115049924A (en) * | 2022-06-06 | 2022-09-13 | 四川大学 | Building earthquake damage assessment method based on non-structural component damage identification under video monitoring |
| CN115082526B (en) * | 2022-07-26 | 2023-02-03 | 复亚智能科技(太仓)有限公司 | Target tracking method and device |
| CN115082526A (en) * | 2022-07-26 | 2022-09-20 | 复亚智能科技(太仓)有限公司 | Target tracking method and device |
| CN116069801A (en) * | 2023-03-06 | 2023-05-05 | 山东华夏高科信息股份有限公司 | Traffic video structured data generation method, device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110516556B (en) | 2023-10-31 |
| WO2021017291A1 (en) | 2021-02-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110516556A (en) | Multi-target tracking detection method, device and storage medium based on Darkflow-DeepSort | |
| Ciaparrone et al. | Deep learning in video multi-object tracking: A survey | |
| Lin et al. | Bsn: Boundary sensitive network for temporal action proposal generation | |
| CN112639873A (en) | Multi-object pose tracking device and method based on single-object pose estimator | |
| CN110555428B (en) | Pedestrian re-identification method, device, server and storage medium | |
| Ren et al. | Learning with weak supervision from physics and data-driven constraints | |
| CN112651995A (en) | On-line multi-target tracking method based on multifunctional aggregation and tracking simulation training | |
| CN110688940A (en) | Rapid face tracking method based on face detection | |
| Jie et al. | Anytime recognition with routing convolutional networks | |
| CN114581954A (en) | Cross-domain retrieval and target tracking method based on pedestrian features | |
| Weng et al. | Whose track is it anyway? improving robustness to tracking errors with affinity-based trajectory prediction | |
| Bashar et al. | Multiple object tracking in recent times: A literature review | |
| CN110991278A (en) | Human body action recognition method and device in video of computer vision system | |
| Ali et al. | Deep Learning Algorithms for Human Fighting Action Recognition. | |
| US20220318555A1 (en) | Action recognition using limited data | |
| CN116883457B (en) | Light multi-target tracking method based on detection tracking joint network and mixed density network | |
| Mittal et al. | Pedestrian detection and tracking using deformable part models and Kalman filtering | |
| CN113313733A (en) | Hierarchical unmanned aerial vehicle target tracking method based on shared convolution | |
| Xue et al. | Tracking multiple visual targets via particle-based belief propagation | |
| Altundogan et al. | Multiple object tracking with dynamic fuzzy cognitive maps using deep learning | |
| CN116129523A (en) | Action recognition method, device, terminal and computer readable storage medium | |
| CN115147385A (en) | Intelligent detection and judgment method for repeated damage in aviation hole exploration video | |
| Di Lascio et al. | Tracking interacting objects in complex situations by using contextual reasoning | |
| Gao et al. | Lightweight high-precision pedestrian tracking algorithm in complex occlusion scenarios. | |
| Patton et al. | Predicting nba talent from enormous amounts of college basketball tracking data |
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 |