CN105513349A - Double-perspective learning-based mountainous area highway vehicle event detection method - Google Patents
Double-perspective learning-based mountainous area highway vehicle event detection method Download PDFInfo
- Publication number
- CN105513349A CN105513349A CN201510854892.6A CN201510854892A CN105513349A CN 105513349 A CN105513349 A CN 105513349A CN 201510854892 A CN201510854892 A CN 201510854892A CN 105513349 A CN105513349 A CN 105513349A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- traffic
- visual angle
- events
- double
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a double-perspective learning-based mountainous area highway vehicle event detection method used for intelligent traffic monitoring. According to the double-perspective learning-based mountainous area highway vehicle event detection method, based on two mutually-independent perspectives, namely, moving object spatial-temporal trajectory mode learning and epipolar plane map-based vehicle moving state analysis, respective independent behaviors of vehicles and overall characteristics of traffic section vehicle flow are detected; the two kinds of characteristics are utilized to detect events such as traffic accidents, traffic congestion, vehicle retrograde travelling and illegal parking; decision level fusion judgment is carried out through correlation processing; a joint inference result can be obtained; and therefore, robustness detection of mountainous area highway vehicle events under a double-perspective fusion framework can be completed. As indicated by experiment results, the detection method has the advantages of low complexity, high real-time performance and excellent performance in terms of a variety of traffic events.
Description
Technical field
The invention belongs to intelligent traffic monitoring and the traffic incidents detection technical field based on image procossing, be specifically related to a kind of mountainous area highway vehicular events detection method based on Double-visual angle study.
Background technology
Along with China's western part great development implementation, contact backwoodsman mountainous area highway fast development.Mountainous area highway has and ensures to travel continuously and the large feature of the traffic capacity, and the needs that it meets personnel, goods flows fast, achieve the lifting of traffic integral level, improve the social benefit of traffic.Because mountainous area highway has higher design capacity, also have larger road traffic simulation power, the specific bottleneck road of part mountainous area highway reaches capacity traffic volume gradually, and all kinds of traffic events obviously increases.Simultaneously because the construction condition of mountainous area highway limits, the reasons such as the area, fog-zone that such as large native stone cube (restriction vision field of driver) and local, mountain ridge meteorological condition cause, cause mountainous area highway traffic events to take place frequently, and traffic hazard is more serious.And a lot of location of mountainous area highway does not possess the condition that managerial personnel's website is accredited at scene of setting up due to the restriction of cost.Automatic traffic event detection and identification are important component parts of freeway traffic regulation system, and its real-time and accuracy are directly connected to the effect of Freeway Traffic Control and traffic guidance.Therefore in time, there is the traffic incidents detection technology of more high precision and Geng Qiang Shandong nation property, especially there is importance for mountainous area highway specific road section.
Current traffic events video Automatic Measurement Technique is not only as effective lifting of existing traffic video monitoring system platform feature, and to there is cost high due to traditional Coil Detector technology I&M, affect road life, affect the defects such as traffic, traffic events video Automatic Measurement Technique is extensively implemented in actual applications.For traffic events video Automatic Measurement Technique, USA and Europe days etc., there was the development of nearly 30 years in country, the application system that the maturation of the International or National widespread use of current acquisition is well-known, mainly comprise the Autoscope series of products of the U.S., the citilog system etc. of Belgian Traficon product and France.Although the cycle of home products development and application is also very short, constantly improve with abundant along with the increasingly mature of video traffic detection technique and function thereof, constantly ripe and its superiority of effectiveness is outstanding all the more based on China-made Video testing product technology.There are Qing Hua Ziguang VS3001, Shenzhen Harbin Institute of Technology electronic traffic VTDZ000, VideoTrace, the Shanghai Gao Dewei of Divine Land, Shenzhen traffic system company limited, sky, the Hunan wing, sad sky, Xiamen, Wenan, Beijing, Chengdu Wei Lute, Astronavigation Age in the producer that traffic events intelligent vision detects independent research at home automatically, move and look unit etc.Current traffic incidents detection properties of product also need improve verification and measurement ratio further and reduce false drop rate and loss, more effectively to promote freeway traffic detection system.Based on certain technical merit widely, verification and measurement ratio and false drop rate two indices conflict mutually, continue to improve verification and measurement ratio when being difficult to ensure false drop rate reduced levels, need to study new technological means, rise to new level with actualizing technology.
Freeway traffic event video Automatic Measurement Technique itself is a kind of content-based image examination and analysb.The fields such as same computer vision, pattern-recognition and information retrieval are the same, and various freeway traffic event all belongs to the content of high-level semantic, inevitably there is semantic gap problem, fully can not be expressed exactly by simple several low-level image feature.Existing great amount of images classification and medical image test experience prove, due to restrictions such as the specificity of single features and the sample dependences of sorter itself, multiple feature and the result of decision thereof are combined, often can obtain performance more better than single sorter.At present commonly use based on vehicle movement trajectory extraction traffic events feature, just an aspect embodying of traffic video content traffic events itself, should obtain more traffic video feature, thus the performance of effective raising traffic incidents detection method.
Traffic congestion, vehicle drive in the wrong direction, rule-breaking vehicle stops and the vehicular events such as traffic hazard be the result of single unit vehicle independent behavior separately, be also the reflection of traffic section wagon flow global feature.In the visual field of video camera, based on the automatic incident detection (AutomaticIncidentDetection of vehicle behavioural characteristic, AID) method can analyze vehicle behavior, the most direct event information is provided, AID method based on wagon flow global feature can obtain the overall wagon flow data of road, and evaluate events affects.For vehicular events, the AID based on vehicle behavioural characteristic and AID two class methods based on wagon flow global feature have its advantages and disadvantages, and a kind of desirable treating method is combined the two, complements one another.The thinking of the general main employing single-view of current traffic incidents detection product, carries out treatment and analyses for image single features.On the one hand, the robustness of single features needs to be improved further for the outdoor environment of various complexity; On the other hand, the essence that the high-level semantic for traffic incidents detection itself is understood, needs feature from various visual angles to complete the contact between low-level image feature to high-level semantic.
Summary of the invention
The present invention is directed to the characteristics of demand of the requirement of mountainous area highway traffic video event detection technology more high precision and more strong robustness, propose a kind of mountainous area highway traffic incidents detection method of robust based on Double-visual angle Learning Principle.
Particularly, the invention provides a kind of mountainous area highway vehicular events detection method based on Double-visual angle study, comprise the following steps:
Step 1, visual angle one: moving target space-time trajectory model learns, and comprises step 1.1 ~ step 1.3.
Step 1.1, traffic video Initialize installation, comprises setting lane line and perform region.
Step 1.2, the moving vehicle based on background modeling detects and track following.
Step 1.3, based on the traffic events identification of vehicle movement trajectory model study.
Step 2, visual angle two: based on the vehicle movement Study on Trend of outer pole-face figure, comprise step 2.1 ~ step 2.3.
Step 2.1, the highway track based on Hough transform and scene dynamics figure is detected automatically.
Step 2.2, the outer pole-face figure of the divided lane moving vehicle space-time based on tracker wire generates.
Lane line according to extracting is the tracker wire that each divided lane arranges through camera coverage scope, ensures that vehicle drives through any time within the scope of camera coverage, all must by a certain bar tracker wire.Divided lane vehicle tracking line can be chosen to be lane line, also can be chosen to be track center line.Respectively the pixel in every bar tracker wire is accumulated along time shaft, generate outer pole-face figure.
Step 2.3, based on the vehicle movement situation feature extraction of the outer pole-face figure of divided lane.
Calculate the slope of outer pole-face figure institute Checking line, obtain Vehicle Speed, and try to achieve vehicle heading and vehicle traveling acceleration.Extract wagon flow global feature by outer pole-face figure, detect traffic congestion and traffic hazard according to space mean speed, whether the angle detecting vehicle according to Vehicle Speed drives in the wrong direction, and space velocity amplitude of variation is for detecting traffic congestion and traffic hazard event.
Whether step 3, based on the Decision-level fusion of Double-visual angle study, occur to detect traffic events, when sending traffic events, and the image-region at traffic events place; Comprise step 3.1 ~ step 3.2.
Step 3.1, merges and passes judgment on.
After identification is made to traffic incidents detection target in each visual angle, the testing result at two visual angles is carried out Decision-level fusion, obtain final decision;
Step 3.2, carries out image co-registration location.
If the rectangle frame in traffic events generation position in the picture represents, this rectangle frame is expressed by four-tuple (x, y, w, h), and (x, y) is the image coordinate in the rectangle frame upper left corner, w and h is respectively the wide and high of rectangle frame;
If the rectangle frame detecting traffic events generation position under a jth visual angle is (x
j, y
j, w
j, h
j), j=1,2; What various visual angles traffic events image co-registration orientation problem is reduced to multiple plane quadrilateral asks Union of Sets Problem, and required polygon is exactly the fusion positioning result of traffic events place image-region.
Compared with prior art, the present invention has following clear superiority:
(1) the present invention is directed to freeway traffic structures and propose a kind of new traffic events video features---based on the outer pole-face figure (EpipolarPlaneImage of divided lane, be called for short EPI) line segment feature slope and angle detecting result, reflect traffic section wagon flow global feature with this;
(2) the present invention is based on Double-visual angle Learning Principle, outer for divided lane pole-face figure is carried out Decision-level fusion with the driving trace feature of the reflection vehicle independent behavior of existing widespread use, realize the detection of vehicle on highway event robust.
(3) experimental result fully shows that algorithm complexity of the present invention is low, and real-time is good, has higher performance for multiple vehicular traffic event.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the mountainous area highway vehicular events detection method that the present invention is based on Double-visual angle study;
Fig. 2 is traffic video Initialize installation schematic diagram, and wherein (a) is original image, and (b) is perform region setting schematic diagram;
Fig. 3 detects schematic diagram automatically based on the highway track of Hough transform and scene dynamics figure; Wherein, (a) is the lane detection result schematic diagram based on Hough transform; B () arranges result schematic diagram for divided lane vehicle tracking line;
Fig. 4 is that outside the divided lane based on tracker wire, pole-face figure generates schematic diagram; A () is for corresponding to the outer pole-face figure of several 4th article of tracker wire from left to right; B () answers the picture frame of EPI tri-lines for traffic video sequence pair;
Fig. 5 is that outer pole-face figure perspective projection correction and line segment feature detect; A () generates result for outer pole-face figure; B () is outer pole-face figure perspective projection correction result; C () is line segment feature testing result;
Fig. 6 is the vehicle congestion testing result based on Double-visual angle; (a1) be vehicle congestion original video frame; (a2) be vehicle congestion vehicle movement track following and event detection outcome; (b1) be the outer pole-face figure of vehicle congestion divided lane; (b2) be line segment feature and the event detection outcome of outer pole-face figure; (c1) be original vehicle tailback occurrence diagram picture; (c2) be the vehicle congestion testing result learnt based on Double-visual angle;
Fig. 7 is the traffic hazard testing result based on Double-visual angle; (a1) be traffic hazard original video frame; (a2) be traffic hazard vehicle movement track following and event detection outcome; (b1) be the outer pole-face figure of traffic hazard divided lane; (b2) be line segment feature and the event detection outcome of outer pole-face figure; (c1) be original traffic accident vehicle occurrence diagram picture; (c2) be the traffic hazard testing result learnt based on Double-visual angle;
Fig. 8 to drive in the wrong direction testing result based on the vehicle of Double-visual angle; (a1) be that vehicle drives in the wrong direction original video frame; (a2) be that vehicle drives in the wrong direction vehicle movement track following and event detection outcome; (b1) be that vehicle drives in the wrong direction the outer pole-face figure of divided lane; (b2) be line segment feature and the event detection outcome of outer pole-face figure; (c1) be that original vehicle is driven in the wrong direction vehicular events image; (c2) be to drive in the wrong direction testing result based on the vehicle of Double-visual angle study;
Fig. 9 is the illegal parking testing result based on Double-visual angle; (a1) be illegal parking original video frame; (a2) be illegal parking vehicle movement track following and event detection outcome; (b1) be the outer pole-face figure of illegal parking divided lane; (b2) be line segment feature and the event detection outcome of outer pole-face figure; (c1) be original illegal parking vehicular events image; (c2) be the illegal parking testing result learnt based on Double-visual angle.
Embodiment
Below in conjunction with drawings and Examples, the present invention will be further described.
The present invention is in the vehicle driving trace feature base of the reflection vehicle independent behavior of existing widespread use, propose to reflect the outer pole-face figure (EpipolarPlaneImage of the divided lane space-time of wagon flow global feature pointedly based on the structurized feature of road traffic, be called for short EPI), thus forming the separate visual angle of traffic events two: vehicle is independent behavior and traffic section wagon flow global feature separately, completes mountainous area highway vehicular events robust and detect under Double-visual angle merges framework.For inputted road monitoring traffic video, for visual angle one, the automatic detection and tracking of moving vehicle are needed, generate the vehicle movement individual behavior feature of vehicle movement real trace data also corresponding to extraction, obtain the concrete detection of traffic events eventually through the study of moving target space-time trajectory model; For visual angle two, need the lane line automatically detecting freeway traffic scene, and generate the outer pole-face figure of space-time of divided lane with this and realize the detection of line segment, the extraction of the wagon flow global feature of the reflection such as final implementation space car speed, acceleration and car speed slope direction section vehicle movement Study on Trend.Utilize that this two category feature detects traffic hazard respectively, traffic congestion, vehicle drive in the wrong direction and the event such as illegal parking, and for the 4 class traffic incidents detection preliminary conclusions that every category feature is set up, Decision-level fusion judgement is carried out, final acquisition associating inferred results by association process.Experimental result shows, detection method complicacy proposed by the invention is low, and real-time is good, the performance that Yan Douyou is higher for multiple traffic events.
The embodiment of the present invention realizes on the PC installing VC2008 and OpenCV2.4.5.
The source of the present embodiment data information is mainly divided into two parts.A part is from domestic multiple high speed monitoring and scheduling command centre.This partial data includes the video monitoring record of traffic hazard that these high speed administrative authoritys occur for system-wide section and traffic events, includes 165 traffic events.Another part data obtained from network opening resource, such as network famous video resource website youku etc.This partial data finally passes judgment on whether suitable applications by factors such as the artificial length according to video data and video capture conditions.This partial data has 478 sections of videos, and major part is all duration about two minutes.This experimental data data finally amounts to 643 traffic events video files, and the traffic video amounting to 1470 minutes is tested.
The flow process that the present invention is based on the mountainous area highway vehicular events detection method of Double-visual angle study as shown in Figure 1, comprises the following steps 1 ~ step 3.
Step 1, one of Double-visual angle feature extraction---moving target space-time trajectory model learns.
Step 1.1, traffic video Initialize installation.
Mainly comprise following content:
1) lane line setting, mainly refers to the lane line position delimiting cross section, section, to plan the traveling-position that vehicle is correct and direction etc., for the generation and position thereof of determining traffic hazard provide reference data;
2) setting of perform region.Setting perform region is to improve computing velocity, removes the calculating in unnecessary region, shields diervilla versicolor, grass, trees simultaneously and shakes impact on accuracy of detection with the wind.
As shown in Figure 2, (a) is the original image in section to be detected, and (b) carries out the image after Initialize installation to original image.Can find out from (b) of Fig. 2, the lane line, detection line etc. of setting.
Step 1.2, based on the moving vehicles detection and tracking of background modeling.
Based on background modeling vehicle detection and follow the tracks of very ripe, be widely used in traffic events video automatic testing method.The object of the method from continuous print traffic video image, extracts moving vehicle target and carries out continuous print recognition and tracking, obtains the information such as the driving trace of each car with this, provides support for successor detects.The embodiment of the present invention just completes the automatic detection and tracking of moving vehicle in the processing procedure of visual angle one based on the ripe algorithm at existing background modeling and regional aim center.
The specific works process of step 1.2 is:
Step 1.2.1, background obtains and upgrades.Generally, always have operational vehicle in video image road obtained to exist.In order to process video image, first need to extract road background, namely eliminate the vehicle that road moves.Background obtains and employs simple background model, namely carries out medium filtering to the frame number in certain hour and obtains background.Context update formula is:
In formula, B
t(x, y) represents the pixel background gray levels of t in (x, y) position, B
t,n(x, y) represents B
tvalue after (x, y) renewal; B
t-1(x, y) is engraved in the pixel background gray levels of (x, y) position when representing t-1; I
t(x, y) represents the gray-scale value of the image of t in (x, y) position.
Step 1.2.2, produces difference image.In order to extract the vehicle target of motion from road traffic operation conditions video image, first needing to carry out difference processing to the background image of current frame image and detection zone, extracting foreground information.T background difference formula is:
D
t(x,y)＝I
t(x,y)-B
t,n(x,y)
In formula, D
t(x, y) represents the difference value of t in (x, y) position.Again to the difference image D obtained
tdo binary conversion treatment, if threshold value is θ, the output image that binary conversion treatment obtains is G
t, wherein G
t(x, y) represents the binary value in (x, y) position.
Then adopt the method that connected region detects, each connected region of statistics prospect, these regions comprise noise and all vehicle targets; Remove noise by the method for shape filtering, produce the image only having prospect (moving object).
Step 1.2.3, determines that target morphology feature is gone forward side by side line trace.The present invention adopts the tracking based on regional aim center to carry out vehicle tracking, namely in vehicle detection result, pixel connected region is one by one identified, represent the vehicle detected with rectangular area, and the morphological feature extracting moving vehicle target area is expressed to vehicle.The embodiment of the present invention specifically uses eccentricity vector to characterize:
According to the connection characteristic of foreground area, region contour point set is sorted counterclockwise and is expressed as a vectorial form: P
t(p
1, p
2..., p
n), n is the pixel number in region contour, element p
irepresent i-th point in region, the coordinate of this point is set to (x
i, y
i).
First, the coordinate (x of zoning center of gravity C
c, y
c):
Then, each element p in profile vector is determined
ieccentricity d
i, eccentricity d
ibe defined as element p
iand the distance between regional barycenter C.Distance thus in vector between each element and center of gravity can form an eccentricity sequence, is the eccentricity vector in foreground target region: D (d by this eccentricity sequence definition
1, d
2..., d
n).
Eccentricity vector average M
1can be expressed as:
Eccentricity vector fractional integration series divergence M
2can be expressed as:
The ratio M of minimax eccentricity
3can be expressed as:
The feature M based on eccentricity vector calculated
1, M
2, M
3in the motion process of regional aim, there is good stability, and there is translation, flexible, invariable rotary characteristic, therefore, by M
1, M
2and M
3as the final expression of target morphology feature.On the basis of moving vehicles detection and tracking, the true driving trace of final each car of acquisition.
Step 1.3, based on the traffic events identification of vehicle movement trajectory model study.
Key factor is the analysis of the vehicle behavioural characteristic relevant to traffic events:
(1) changes in vehicle speed: if most vehicle reduces suddenly in certain region progresses of arrival speed of sailing, and increase gradually after by this region, traffic events so may be had on this region to occur.
(2) vehicle lay-off: if vehicle unexpected stagnation of movement on highway, be commonly referred to be due to automobile crash, vehicle cast anchor, vehicle trouble or other unpredicted reason cause.
(3) vehicle crossover lane and direction: if all vehicles are all changing track or travel direction through certain section of region, illustrate that this track or front have accident to occur, vehicle afterwards must be kept away barrier and travel.
Step 2, based on Double-visual angle feature extraction two---based on the vehicle movement Study on Trend of outer pole-face figure.
The thinking of the general main employing single-view of the current automatic testing product of traffic events video, carries out treatment and analyses for moving vehicle space-time track single features.On the one hand, the robustness of moving vehicle space-time track following needs to be improved further for the outdoor environment of various complexity; On the other hand, the essence that the high-level semantic for traffic incidents detection itself is understood, needs feature from various visual angles to complete the contact between low-level image feature to high-level semantic.
Step 2.1, the track, expressway based on Hough transform and scene dynamics figure is detected automatically.
Due to the scene that Expressway Road is specific highly structural, vehicle normally can only fix traveling along track.In the specific scene of highway, the information of lane line has showed the normal transport condition of vehicle completely.Therefore, the first step of this work extracts lane line information exactly, as shown in Figure 3.In (a) and (b) of Fig. 3, black line is the lane line extracted.
Step 2.2, the outer pole-face figure of the divided lane moving vehicle space-time based on tracker wire generates.
According to the lane line that step 2.1 is extracted, be respectively each track and the tracker wire running through whole camera coverage scope is set.Vehicle tracking line can be chosen to be lane line, also can be chosen to be track center line, i.e. the centre position of two adjacent lane lines, as long as ensure that each car drives through any time within the scope of camera coverage, and all must by a certain bar tracker wire.Respectively these tracker wires are accumulated on a timeline again, generate outer pole-face figure (EpipolarPlaneImage, EPI).Therefore, by processing procedure like this, motion analysis can be converted into the extraction of trajectory on two-dimentional EPI.
The embodiment of the present invention chooses lane line or/and the medium line of adjacent lane line is tracker wire.Fig. 4 illustrates the generative process of outer pole-face, and wherein (a) is the generated outer pole-face figure corresponding to a certain bar tracker wire, and horizontal ordinate illustrates the spatial information of tracker wire, and ordinate illustrates time frame information.A in (), three black lines represent that different time chart picture frame arranges along the pixel of this tracker wire.(b) of Fig. 4 traffic video sequence corresponding to tracker wire not picture frame in the same time, black line wherein represents the set tracker wire due to many tracks.
(a) ~ (c) of Fig. 5 illustrates the process of outer pole-face figure perspective projection correction and Line segment detection, and detected line segment is as shown in white line in figure.
Step 2.3, based on the vehicle movement situation feature extraction of the outer pole-face figure of divided lane.
The analysis of traffic section entire vehicle situation is just converted into extraction and the slope analysis thereof of line segment feature on EPI image, as shown in Figure 5.Calculate the slope that the outer pole-face figure institute of all vehicles forms Line segment detection, be the expression of Vehicle Speed, and try to achieve direction and the vehicle traveling acceleration of Vehicle Speed further.Key factor is the calculating of the wagon flow global feature relevant to traffic events:
(1) average spatial velocity, also be track average velocity: after traffic events occurs, having vehicle stops on track, road effective width is reduced, hinder other vehicle normal pass, other vehicle is can reduce by average velocity during spot, i.e. the average spatial velocity of traffic flow can reduce.This feature can detect the larger event of the coverage such as traffic congestion, traffic hazard.
(2) direction of Vehicle Speed: for pole-face figure outside bicycle road, due to the position influence that video camera is supposed, the vehicle of each outer pole-face figure is not from the close-by examples to those far off be exactly on the contrary.The line segment slope direction of these two kinds of directions outside on pole-face figure is contrary.Determine the direction of Vehicle Speed according to the line segment slope direction on outer pole-face figure, this feature can detect vehicle and to drive in the wrong direction event.
(3) space velocity amplitude of variation, refer to the average spatial velocity amplitude of variation of traffic flow: after traffic events occurs, having vehicle stops on track, road effective width and number of track-lines are reduced, hinder other vehicle normal pass, other vehicle is can reduce by average velocity during spot, i.e. the average spatial velocity of traffic flow can reduce.This feature can detect the larger event of the coverage such as traffic congestion, traffic hazard.
Step 3, based on the Decision-level fusion of Double-visual angle study.
The final Output rusults of mountainous area highway traffic incidents detection based on Double-visual angle study is divided into two parts content.First be the judgement of traffic incidents detection, namely whether a certain traffic events occurs.This partial content relates to fusion each visual angle traffic incidents detection result, obtains final fusion evaluation result.Next is traffic incidents detection positioning result, the image-region at namely occurred traffic events place, and this partial content relates to image co-registration location.
Step 3.1, Decision-level fusion is passed judgment on.
For the mountainous area highway traffic incidents detection method based on Double-visual angle study proposed by the invention, the video low-level image feature of visual angle all have expressed traffic events from a concrete side, can complete traffic events video and automatically detect.After identification is made to traffic incidents detection target in each visual angle, based on the traffic incidents detection method of Double-visual angle study, the testing result at two visual angles is carried out Decision-level fusion, finally obtains the testing result that more robust is consistent.
Decision-level fusion refers to after identification is made to traffic incidents detection target in each visual angle, the testing result at multiple visual angle is carried out merging and finally obtains overall consistent decision-making.Decision level fusion is a kind of high-level fusion, directly for concrete decision objective, makes full use of all kinds of characteristic informations that feature-based fusion draws, and provides concisely and intuitively result.Conveniently calculate, often select "AND" criterion and "or" criterion.Wherein, "AND" criterion allows traffic events to be received to real testing result, the result that the decision only provided in the feature that the use of all different subsystems is different is positive.Use the method for this rule to produce a low false acceptance rate, result also in higher false rejection rate.Permission traffic events is received to real testing result, as long as any one subsystem have passed detection by "or" criterion.Generally, this rule can produce a lower false rejection rate and higher false acceptance rate.Select in the embodiment of the present invention "or" criterion complete for Double-visual angle separately recognition detection result Decision-level fusion pass judgment on.
Step 3.2, image co-registration is located.
Traffic events image co-registration location based on various visual angles study needs reasonably to be merged by each visual angle positioning result.Traffic events generation position is in the picture represented by a rectangle frame.If traffic incidents detection locating rectangle frame is expressed by four-tuple (x, y, w, h), represent the image coordinate (x, y) in the traffic events rectangle frame upper left corner and the wide w of rectangle frame and high h respectively.If the traffic incidents detection locating rectangle frame at each visual angle is (x
j, y
j, w
j, h
j), there are two visual angles, then j=1 in the corresponding each visual angle of j, 2 in the present invention.What the traffic events image co-registration orientation problem based on various visual angles study can be reduced to multiple plane quadrilateral asks Union of Sets Problem.Required polygon is exactly the traffic events image co-registration positioning result of various visual angles study.
4 kinds of traffic events that the present invention is correlated with vehicle: traffic congestion, vehicle drives in the wrong direction, rule-breaking vehicle stops, traffic hazard, for example, in the vehicle driving trace feature base of the reflection vehicle independent behavior of existing widespread use, propose the outer pole-face figure of divided lane space-time pointedly based on the structurized feature of road traffic and reflect wagon flow global feature, thus form the feature at traffic events two visual angles (vehicle is independent behavior and traffic section wagon flow global feature separately) itself, under Double-visual angle merges framework, complete mountainous area highway vehicular events robust detect.
The traffic incidents detection of all kinds all needs through these steps, to complete the traffic events video feature extraction of Double-visual angle, and realization fusion judges to draw the judgement whether traffic events of more robust occurs and determine the position that vehicular events occurs further.Analyze mainly for various vehicular traffic events involved in the present invention below.
(1) traffic congestion.
Vehicle movement track represents the time dependent real-time information of vehicle location.By visual angle one, to vehicle movement track following, the shift transformation of each car can be calculated, obtain the travel speed of vehicle in real time.When the travel speed of all vehicles has the trend of reduction, then think that this section there occurs vehicle congestion.Based on the detection method Main Analysis track average velocity trend over time of outer pole-face figure.By visual angle two, obtain track average velocity trend over time according to outer pole-face figure, if the average velocity in all tracks is lower than preset value, just thinks and there occurs vehicle congestion.Based on Double-visual angle study traffic congestion testing process as shown in Figure 6.In Fig. 6, (a1) be vehicle congestion original video, this section is detected by visual angle one, obtain (a2) vehicle congestion vehicle movement track following and event detection outcome, detected by visual angle two, obtain line segment feature and the event detection outcome of pole-face figure, (b2) outer pole-face figure outside (b1) vehicle congestion divided lane.(c1) of Fig. 6 is for the video image of vehicle congestion traffic events being detected; (c2) be the vehicle congestion testing result figure learnt based on Double-visual angle.
(2) traffic hazard.
In the detection method of visual angle one based on vehicle movement track, calculate vehicle spacing change between any two.When there being spacing to be less than given threshold value, then there occurs vehicle collision, and major embodiment is the ANOMALOUS VARIATIONS of track of vehicle.The change of track average velocity is mainly compared based on the detection method of outer pole-face figure in visual angle two.If certain track average velocity amplitude of variation is comparatively large, exceedes setting threshold value, just think and there occurs traffic hazard event.Based on Double-visual angle study traffic hazard testing process as shown in Figure 7.In Fig. 7, (a1) relates to certain frame raw video image of traffic hazard; (a2) be detected by visual angle one, carry out traffic hazard vehicle movement track following and event detection outcome; (b1) be the outer pole-face figure of traffic hazard divided lane obtained by visual angle two detection; (b2) be line segment feature and the event detection outcome of outer pole-face figure.(c1) of Fig. 7 is video image traffic hazard vehicular events being detected, and (c2) is the traffic hazard testing result detected by the inventive method.
(3) vehicle drives in the wrong direction.
Visual angle one calculates the direction vector before and after vehicle between two frames based on the detection method of vehicle movement track, obtains vehicle and to drive in the wrong direction the judgement of behavior.Be normal travel direction f to detect video interested region direction initialization, then before and after normal direction vehicle, the direction vector of two frame center of mass motions should this specific direction all the time.When occurring if drive in the wrong direction, then before and after the movement locus of this retrograde vehicle, the direction vector of two frame center of mass motions is not identical with f.Generally take multiframe to be added and the mode be averaged judges whether to drive in the wrong direction, prevent interference from causing erroneous judgement with this.
All line segments are divided into two classes based on the detection method of outer pole-face figure with velocity reversal by visual angle two.If the line segment contrary with predefined normal travel direction is gathered in the larger scope A of setting, just think that there occurs vehicle drives in the wrong direction event.A is positive number, represents a range size.
Vehicle based on Double-visual angle study drives in the wrong direction testing process as shown in Figure 8.In Fig. 8, (a1) is that vehicle drives in the wrong direction raw video image; (a2) be to be driven in the wrong direction for a pair vehicle movement track following and event detection outcome by visual angle; (b1) be that the vehicle detected by visual angle two is driven in the wrong direction the outer pole-face figure of divided lane; (b2) be line segment feature and the event detection outcome of outer pole-face figure.(c1) of Fig. 8 detects that original vehicle is driven in the wrong direction the video image of vehicular events, and (c2) to drive in the wrong direction result with the vehicle that the inventive method detects.
(4) illegal parking.
The time dependent process in track of vehicle position is analyzed based on the detection method of Vehicle tracing in visual angle one.When rule-breaking vehicle stops, front and back framing bit moves and constantly reduces until finally slack.Visual angle two is based on the detection method Main Analysis track average velocity of outer pole-face figure.The outer pole-face figure generated for bicycle road carries out Line segment detection, relatively track average velocity trend over time, if track average velocity constantly reduces and is tending towards 0, and these velocity variations regions can be gathered within default larger scope, just think be present in rule-breaking vehicle stop traffic events.。Based on Double-visual angle study illegal parking testing process as shown in Figure 9.In Fig. 9, (a1) is illegal parking original video; (a2) be by a pair, visual angle illegal parking vehicle movement track following and event detection outcome; (b1) be the outer pole-face figure of illegal parking divided lane detected by visual angle two; (b2) be line segment feature and the event detection outcome of outer pole-face figure.(c1) of Fig. 9 is the original illegal parking vehicular events video image detected, (c2) is the illegal parking testing result that the inventive method detects.
The present invention specifically adopts the performances of 4 indexs to vehicular events Video Detection Algorithm such as verification and measurement ratio, false drop rate, loss and average detected time to evaluate:
Verification and measurement ratio=TP/P × 100%
False drop rate=FP/N × 100%
Loss=(1-verification and measurement ratio) × 100%
Wherein, P, N are respectively positive and negative sample number, and TP is the sample number that positive pattern detection is correct, and FP is negative sample error detection is correct sample number, and M is the number of detected traffic events, t
athe time that traffic incidents detection method detects event, and t
incit is the time that traffic events occurs.The traffic events belonging to all videos is demarcated by manual type.
The testing result (the previous numeral of every lattice) of the mountainous area highway vehicular events detection method that the present invention is based on Double-visual angle study added up by table 1, and carries out contrasting (after every lattice a numeral) with traditional single-view algorithm (vehicle movement track).
The vehicular events testing result that table 1 learns based on Double-visual angle is added up and contrast table
Test the embodiment of the present invention, the final verification and measurement ratio finally obtaining detection method is 94.09%, and false drop rate is 4.51%, and loss is 1.40%.And the final verification and measurement ratio of single-view algorithm is 87.33%, false drop rate is 6.99%, and loss is 5.68%.The average detected time does not have marked change.Associative list 1 can obtain, and compared with single-view detection algorithm, vehicular events detection method of the present invention is generally all greatly improved for the detection perform of multiple traffic events.
Claims (5)
1., based on a mountainous area highway vehicular events detection method for Double-visual angle study, it is characterized in that, comprise the following steps:
Step 1, visual angle one: moving target space-time trajectory model learns, and comprises step 1.1 ~ step 1.3;
Step 1.1, traffic video Initialize installation, comprises setting lane line and perform region;
Step 1.2, the moving vehicle based on background modeling detects and track following;
Step 1.3, based on the traffic events identification of vehicle movement trajectory model study;
Step 2, visual angle two: based on the vehicle movement Study on Trend of outer pole-face figure, comprise step 2.1 ~ step 2.3;
Step 2.1, the track, expressway based on Hough transform and scene dynamics figure is detected automatically;
Step 2.2, the outer pole-face figure of the divided lane moving vehicle space-time based on tracker wire generates;
Lane line according to extracting is the tracker wire that each divided lane arranges through camera coverage scope, ensures that vehicle drives through any time within the scope of camera coverage, all must by a certain bar tracker wire; Respectively the pixel in every bar tracker wire is accumulated along time shaft, generate outer pole-face figure;
Step 2.3, based on the vehicle movement situation feature extraction of the outer pole-face figure of divided lane;
Calculate the slope of outer pole-face figure institute Checking line, obtain Vehicle Speed, and try to achieve vehicle heading and vehicle traveling acceleration;
Whether step 3, based on the Decision-level fusion of Double-visual angle study, comprises two results: detect traffic events and occur; When there is traffic events, the image-region at traffic events place, location; Comprise step 3.1 ~ step 3.2;
Step 3.1, after identification is made to traffic incidents detection target in each visual angle, carries out Decision-level fusion by the testing result at two visual angles, obtains final decision;
Step 3.2, carries out the warm location of image;
If the rectangle frame in traffic events generation position in the picture represents, this rectangle frame is expressed by four-tuple (x, y, w, h), and (x, y) is the image coordinate in the rectangle frame upper left corner, w and h is respectively the wide and high of rectangle frame;
If the rectangle frame detecting traffic events generation position under a jth visual angle is (x
j, y
j, w
j, h
j), j=1,2; What various visual angles traffic events image co-registration orientation problem is reduced to multiple plane quadrilateral asks Union of Sets Problem, and the polygon of trying to achieve is exactly the fusion positioning result of traffic events place image-region.
2. the mountainous area highway vehicular events detection method based on Double-visual angle study according to claim 1, it is characterized in that, in described step 2.1, the line selection of each divided lane vehicle tracking is decided to be lane line or is chosen to be track center line.
3. the mountainous area highway vehicular events detection method based on Double-visual angle study according to claim 1, is characterized in that, in described step 2.3, extract wagon flow global feature, comprising by outer pole-face figure:
(1) space mean speed: after traffic events occurs, have vehicle to stop on track, road effective width is reduced, hinder other vehicle normal pass, other vehicle average velocity when passing through spot will reduce; Space mean speed is for detecting traffic congestion and traffic hazard;
(2) whether the direction of Vehicle Speed, determines according to the line segment slope direction on outer pole-face figure, drive in the wrong direction for detecting vehicle;
(3) space velocity amplitude of variation: the average spatial velocity amplitude of variation referring to traffic flow; After traffic events occurs, have vehicle to stop on track, road effective width and number of track-lines are reduced, hinders other vehicle normal pass, other vehicle average velocity when passing through spot reduces; Space velocity amplitude of variation is for detecting traffic congestion and traffic hazard event.
4. the mountainous area highway vehicular events detection method based on Double-visual angle study according to claim 1, is characterized in that, in described step 3, based on traffic incidents detection several below Double-visual angle decision-making;
(1) traffic congestion;
By visual angle one, to vehicle movement track following, calculate in real time the travel speed obtaining vehicle, when the travel speed of all vehicles has the trend of reduction, then think that this section there occurs vehicle congestion;
By visual angle two, according to the average velocity trend over time that outer pole-face figure obtains track, if all tracks average velocity is lower than preset value, then thinks and there occurs vehicle congestion;
(2) traffic hazard;
By visual angle one, calculating vehicle spacing change between any two, when there being spacing to be less than given threshold value, then there occurs vehicle collision;
By visual angle two, compare the change of track average velocity, if certain track average velocity amplitude of variation exceedes setting threshold value, then think and there occurs traffic hazard event;
(3) vehicle drives in the wrong direction;
By visual angle one, calculate the direction vector between two frames before and after vehicle, obtain vehicle and to drive in the wrong direction the judgement of behavior;
By visual angle two, with velocity reversal, all line segments are divided into two classes, if the line segment contrary with predefined normal vehicle operation direction be gathered in set size scope in, then think that there occurs vehicle drives in the wrong direction event;
(4) illegal parking;
By visual angle one, to vehicle movement track following, when rule-breaking vehicle stops, front and back framing bit moves and constantly reduces until finally slack;
By visual angle two, the outer pole-face figure generated for bicycle road carries out Line segment detection, relatively track average velocity trend over time, if track average velocity constantly reduces and is tending towards 0, and the region clustering of velocity variations is within the scope of default size, just think be present in rule-breaking vehicle stop traffic events.
5. the mountainous area highway vehicular events detection method based on Double-visual angle study according to claim 4, it is characterized in that, in described step 3, to the testing result at visual angle one and visual angle two, "AND" criterion or "or" criterion is adopted to carry out the last testing result of decision-making.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510854892.6A CN105513349B (en) | 2014-12-03 | 2015-11-30 | Mountainous area highway vehicular events detection method based on double-visual angle study |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410718808 | 2014-12-03 | ||
| CN2014107188083 | 2014-12-03 | ||
| CN201510854892.6A CN105513349B (en) | 2014-12-03 | 2015-11-30 | Mountainous area highway vehicular events detection method based on double-visual angle study |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105513349A true CN105513349A (en) | 2016-04-20 |
| CN105513349B CN105513349B (en) | 2017-12-05 |
Family
ID=55721290
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510854892.6A Active CN105513349B (en) | 2014-12-03 | 2015-11-30 | Mountainous area highway vehicular events detection method based on double-visual angle study |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105513349B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105741565A (en) * | 2016-04-21 | 2016-07-06 | 正元地理信息有限责任公司 | Method for automatic determining accident and giving off alarm based on monitoring video |
| CN106652445A (en) * | 2016-11-15 | 2017-05-10 | 成都通甲优博科技有限责任公司 | Road traffic accident judging method and device |
| CN107293116A (en) * | 2017-06-27 | 2017-10-24 | 浙江浩腾电子科技股份有限公司 | A kind of traffic incident detecting system based on video analysis |
| CN108898844A (en) * | 2018-07-18 | 2018-11-27 | 希社（上海）智能交通科技有限公司 | Suspicion evidence taking system for illegal parking and method |
| CN110264577A (en) * | 2019-06-26 | 2019-09-20 | 中国人民解放军火箭军工程大学 | A kind of collision real-time detection method based on temporal and spatial correlations tracking strategy |
| CN113436434A (en) * | 2021-06-25 | 2021-09-24 | 中科路恒工程设计有限公司 | Mountain trunk highway high-risk road section early warning system and method |
| CN113936465A (en) * | 2021-10-26 | 2022-01-14 | 公安部道路交通安全研究中心 | Traffic incident detection method and device |
| CN114333330A (en) * | 2022-01-27 | 2022-04-12 | 浙江嘉兴数字城市实验室有限公司 | Intersection event detection system and method based on roadside edge holographic sensing |
| CN114882393A (en) * | 2022-03-29 | 2022-08-09 | 华南理工大学 | Road reverse running and traffic accident event detection method based on target detection |
| CN114944062A (en) * | 2022-05-30 | 2022-08-26 | 长安大学 | Construction method of tunnel parallel traffic system |
| CN115148018A (en) * | 2021-03-31 | 2022-10-04 | 海信集团控股股份有限公司 | Traffic incident detection apparatus and method |
| CN117095540A (en) * | 2023-10-18 | 2023-11-21 | 四川数字交通科技股份有限公司 | Early warning method and device for secondary road accidents, electronic equipment and storage medium |
| CN117636270A (en) * | 2024-01-23 | 2024-03-01 | 南京理工大学 | Vehicle robbery event identification method and device based on monocular camera |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080221783A1 (en) * | 2006-04-18 | 2008-09-11 | Gregory Jensen Boss | Intelligent redirection of vehicular traffic due to congestion and real time performance metrics |
| CN101751782A (en) * | 2009-12-30 | 2010-06-23 | 北京大学深圳研究生院 | Crossroad traffic event automatic detection system based on multi-source information fusion |
| WO2011126215A2 (en) * | 2010-04-09 | 2011-10-13 | 고려대학교 산학협력단 | Traffic flow control and dynamic path providing system linked with real-time traffic network structure control based on bidirectional communication function-combined vehicle navigation, and method thereof |
| CN103985250A (en) * | 2014-04-04 | 2014-08-13 | 浙江工业大学 | Light-weight holographic road traffic state visual inspection device |
| CN104077913A (en) * | 2013-03-27 | 2014-10-01 | 上海市城市建设设计研究总院 | Multi-view image information-fused traffic accident monitoring method and device |
-
2015
- 2015-11-30 CN CN201510854892.6A patent/CN105513349B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080221783A1 (en) * | 2006-04-18 | 2008-09-11 | Gregory Jensen Boss | Intelligent redirection of vehicular traffic due to congestion and real time performance metrics |
| CN101751782A (en) * | 2009-12-30 | 2010-06-23 | 北京大学深圳研究生院 | Crossroad traffic event automatic detection system based on multi-source information fusion |
| WO2011126215A2 (en) * | 2010-04-09 | 2011-10-13 | 고려대학교 산학협력단 | Traffic flow control and dynamic path providing system linked with real-time traffic network structure control based on bidirectional communication function-combined vehicle navigation, and method thereof |
| CN104077913A (en) * | 2013-03-27 | 2014-10-01 | 上海市城市建设设计研究总院 | Multi-view image information-fused traffic accident monitoring method and device |
| CN103985250A (en) * | 2014-04-04 | 2014-08-13 | 浙江工业大学 | Light-weight holographic road traffic state visual inspection device |
Non-Patent Citations (2)
| Title |
|---|
| 叶荣炬 等: "1种基于车辆时空图的车辆异常行为检测方法", 《交通信息与安全》 * |
| 赵有婷 等: "基于视频车辆轨迹模型的交通事件自动检测方法研究", 《中山大学学报（自然科学版）》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105741565A (en) * | 2016-04-21 | 2016-07-06 | 正元地理信息有限责任公司 | Method for automatic determining accident and giving off alarm based on monitoring video |
| CN106652445A (en) * | 2016-11-15 | 2017-05-10 | 成都通甲优博科技有限责任公司 | Road traffic accident judging method and device |
| CN106652445B (en) * | 2016-11-15 | 2019-08-23 | 成都通甲优博科技有限责任公司 | A kind of road traffic accident method of discrimination and device |
| CN107293116A (en) * | 2017-06-27 | 2017-10-24 | 浙江浩腾电子科技股份有限公司 | A kind of traffic incident detecting system based on video analysis |
| CN108898844A (en) * | 2018-07-18 | 2018-11-27 | 希社（上海）智能交通科技有限公司 | Suspicion evidence taking system for illegal parking and method |
| CN110264577A (en) * | 2019-06-26 | 2019-09-20 | 中国人民解放军火箭军工程大学 | A kind of collision real-time detection method based on temporal and spatial correlations tracking strategy |
| CN110264577B (en) * | 2019-06-26 | 2020-04-17 | 中国人民解放军火箭军工程大学 | Real-time collision detection method based on space-time correlation tracking strategy |
| CN115148018B (en) * | 2021-03-31 | 2024-05-24 | 海信集团控股股份有限公司 | Traffic event detection device and method |
| CN115148018A (en) * | 2021-03-31 | 2022-10-04 | 海信集团控股股份有限公司 | Traffic incident detection apparatus and method |
| CN113436434B (en) * | 2021-06-25 | 2022-05-27 | 中科路恒工程设计有限公司 | Mountain trunk highway high-risk road section early warning system and method |
| CN113436434A (en) * | 2021-06-25 | 2021-09-24 | 中科路恒工程设计有限公司 | Mountain trunk highway high-risk road section early warning system and method |
| CN113936465A (en) * | 2021-10-26 | 2022-01-14 | 公安部道路交通安全研究中心 | Traffic incident detection method and device |
| CN113936465B (en) * | 2021-10-26 | 2023-08-18 | 公安部道路交通安全研究中心 | Traffic event detection method and device |
| CN114333330A (en) * | 2022-01-27 | 2022-04-12 | 浙江嘉兴数字城市实验室有限公司 | Intersection event detection system and method based on roadside edge holographic sensing |
| CN114333330B (en) * | 2022-01-27 | 2023-04-25 | 浙江嘉兴数字城市实验室有限公司 | Intersection event detection system based on road side edge holographic sensing |
| CN114882393A (en) * | 2022-03-29 | 2022-08-09 | 华南理工大学 | Road reverse running and traffic accident event detection method based on target detection |
| CN114944062A (en) * | 2022-05-30 | 2022-08-26 | 长安大学 | Construction method of tunnel parallel traffic system |
| CN117095540A (en) * | 2023-10-18 | 2023-11-21 | 四川数字交通科技股份有限公司 | Early warning method and device for secondary road accidents, electronic equipment and storage medium |
| CN117095540B (en) * | 2023-10-18 | 2024-01-23 | 四川数字交通科技股份有限公司 | Early warning method and device for secondary road accidents, electronic equipment and storage medium |
| CN117636270A (en) * | 2024-01-23 | 2024-03-01 | 南京理工大学 | Vehicle robbery event identification method and device based on monocular camera |
| CN117636270B (en) * | 2024-01-23 | 2024-04-09 | 南京理工大学 | Vehicle robbery event identification method and device based on monocular camera |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105513349B (en) | 2017-12-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105513349A (en) | Double-perspective learning-based mountainous area highway vehicle event detection method | |
| CN110472496B (en) | Traffic video intelligent analysis method based on target detection and tracking | |
| CN103021186B (en) | Vehicle monitoring method and vehicle monitoring system | |
| CN112700470B (en) | Target detection and track extraction method based on traffic video stream | |
| CN104537841B (en) | Unlicensed vehicle violation detection method and detection system thereof | |
| CN103116987B (en) | Traffic flow statistic and violation detection method based on surveillance video processing | |
| Lin et al. | A Real‐Time Vehicle Counting, Speed Estimation, and Classification System Based on Virtual Detection Zone and YOLO | |
| CN108230254A (en) | A kind of full lane line automatic testing method of the high-speed transit of adaptive scene switching | |
| CN105744232A (en) | Method for preventing power transmission line from being externally broken through video based on behaviour analysis technology | |
| CN103235933A (en) | Vehicle abnormal behavior detection method based on Hidden Markov Model | |
| US20180089515A1 (en) | Identification and classification of traffic conflicts using live video images | |
| CN103377555A (en) | Method and system for automatically detecting anomalies at a traffic intersection | |
| CN107886055A (en) | A kind of retrograde detection method judged for direction of vehicle movement | |
| CN110688922A (en) | Deep learning-based traffic jam detection system and detection method | |
| CN105424330A (en) | Lorry shaft end bolt fault detecting method and lorry shaft end bolt fault detecting device | |
| CN105575125A (en) | Vehicle flow video detection and analysis system | |
| EP2813973B1 (en) | Method and system for processing video image | |
| CN104200466A (en) | Early warning method and camera | |
| CN114781479A (en) | Traffic incident detection method and device | |
| CN113936465B (en) | Traffic event detection method and device | |
| US20220237919A1 (en) | Method, Apparatus, and Computing Device for Lane Recognition | |
| CN106204650A (en) | A kind of vehicle target tracking based on vacant lot video corresponding technology | |
| Meng et al. | Video‐Based Vehicle Counting for Expressway: A Novel Approach Based on Vehicle Detection and Correlation‐Matched Tracking Using Image Data from PTZ Cameras | |
| CN111160132B (en) | Method and device for determining lane where obstacle is located, electronic equipment and storage medium | |
| Makhmutova et al. | Intelligent detection of object's anomalies for road surveilance cameras |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |