CN103440771A - Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection - Google Patents
Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection Download PDFInfo
- Publication number
- CN103440771A CN103440771A CN2013104066535A CN201310406653A CN103440771A CN 103440771 A CN103440771 A CN 103440771A CN 2013104066535 A CN2013104066535 A CN 2013104066535A CN 201310406653 A CN201310406653 A CN 201310406653A CN 103440771 A CN103440771 A CN 103440771A
- Authority
- CN
- China
- Prior art keywords
- car
- car light
- pairing
- vehicle
- light
- 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.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a night complex vehicle detection method. The method comprises: first extracting vehicle lamps by using the homomorphic filtering technology in a frequency domain and the orientation fuzzy technology in a time domain; then performing pairing on the vehicle lamps by using the statistics information of the successfully paired vehicle lamps and introducing a vehicle lamp pairing feedback correction mechanism herein so as to enable a pairing result to be more accurate; and at last tracking vehicles and introducing a vehicle tracking feedback correction mechanism so as to ensure a single corresponding locus for a single vehicle. The shielding problems are respectively handled by using the vehicle lamp pairing feedback correction mechanism for comparison and determination. By the application provided by the invention, higher-accuracy vehicle detection and tracking can be realized, and an overall tracking system has higher precision.
Description
Technical field
What the present invention relates to is a kind of vehicle movement analytical approach, specifically homomorphic filtering azimuth ambiguity technology combines, then with the fuzzy membership of the features such as car, car light is followed the trail of and matched, introduce car light pairing feedback modifiers mechanism makes the pairing result more accurate simultaneously, and solved occlusion issue, finally utilize the pairing car light to realize car tracing, the car tracing feedback modifiers module of herein introducing is guaranteed car one track.The method can realize accurate location and the tracking of driving vehicle in complicated traffic video at night.
Background technology
The vehicle movement analysis is normally followed the trail of target vehicle, obtains the information such as movement velocity, displacement, track of vehicle, for the evidence obtaining of traffic hazard processing, unlawful practice analysis and crime of peace security provides important evidence.We accurately locate and follow the trail of a target vehicle, while particularly vehicle at night being followed the trail of, to avoid the interference of other vehicles, reflected light, street lamp as far as possible, to consider the complicacy of environment simultaneously, as sleet weather, vehicle is in negotiation of bends, traffic congestion etc., this is the difficult point that vehicle at night is followed the trail of, and is also the focus that vehicle at night is studied in following the trail of in recent years.
In complicated traffic video at comparatively ripe night, vehicle checking method mainly divides four classes at present: (1) vehicle detection based on the night vision sensor; (2) vehicle detection based on assisting vehicle; (3) vehicle detection based on coordinate transform; (4) vehicle detection based on car light and windshield feature.At these, in comparatively ripe algorithm, method (1) is subject to impact and the apparatus expensive of intensity of light, and cost is high; Method (2) only can detect the vehicle in the place ahead, the visual field, for the assisting vehicle rear or in assisting vehicle the place ahead vehicle at a distance, the method all can't be detected it, so the method can not be for car tracing; Method (3) is changed world coordinates, camera coordinates and image coordinate, and this needs a large amount of surveying works, and the operational data complexity, and process is loaded down with trivial details.And the vehicle checking method based on car light in method (4), cost is low, and (not needing assisting vehicle) simple to operate, operand is little, and (not being subject to the impact of car light intensity) applied widely, these advantages make it become the focus of broad research in this year.But, current the proposed vehicle checking method based on the car light feature, at complex condition (rain, snow day), bend or car light number are 1 and more than in the situation of 2, follow the trail of result bad.
Night that this patent is mentioned complicated vehicle detection method, that the homomorphic filtering of frequency domain is combined with the azimuthal blur technology of time domain, extract car light, recycle and successfully match the statistical information of car light car light is matched, be incorporated herein car light pairing feedback modifiers mechanism, the result that makes to match is more accurate, finally vehicle is followed the trail of, introduce car tracing feedback modifiers mechanism, guaranteeing to follow the trail of result is car one track simultaneously.The frequency domain homomorphic filtering has strengthened the contrast of image, street lamp, advertising license plate bright lamp have been reduced, car light irradiation is the interference to locate lamp area to other luminous objects of the non-car lightings such as the reflected light of other vehicle body and road reflection light, makes locate lamp area more accurate.The azimuthal blur technology, in the situation that car light and reflected light contrast are lower, still can accurately be extracted car light.For occlusion issue, use car light pairing feedback modifiers mechanism, compare judgement, it is processed respectively.
Summary of the invention
The object of the present invention is to provide the method for vehicle detection and tracking in complicated traffic video at a kind of night that can realize greater efficiency: vehicle detection and method for tracing based on feature Fuzzy degree of membership and feedback modifiers mechanism.
Specific implementation step of the present invention is as follows:
1), figure image intensifying: utilize homomorphic filter, at frequency domain, image is processed, strengthen the contrast of image.
2), car light extracts: to the image after strengthening, application azimuthal blur technology and automatic threshold technology are extracted car light., eliminate because street lamp, billboard, ground return, headlight are got to the interference to locate lamp area such as other vehicle body reflected light, blurring mapping simultaneously, obtain car light position and shape information accurately.
3), the bicycle lamp follows the trail of: in view of car light motion in image is slowly, and interframe has the area of repetition, so locate to introduce this feature of area severe, follows the trail of car light.Caravan lamp area multiplicity is E
oarea, computing formula is as follows:
Wherein, S
overlap(M, i, M+1, j) represents the area of M frame car light i and M+1 frame car light j repeating part, and S (M, i) represents the area of M frame car light i, and S (M+1, j) represents the area of M+1 frame car light j.Two car lights of long-pending multiplicity maximum, be labeled as a car cresset.
4), the car light pairing is feedback modifiers mechanism: the information of the car light of the successful pairing at first we extracted is added up, calculate again the fuzzy membership of the features such as car (distance is than, area difference, angle) according to statistical information, use the feature Fuzzy degree of membership to be matched to car light, introduce pairing simultaneously and revise feedback mechanism, guarantee the accuracy of pairing, also well solved the marriage problem blocked while occurring.Concrete grammar is as follows:
1. for the car light existed in same two field picture, the distance of calculating any two car lights than, the ratio of less ordinate in two car light Diff Es and two car lights, be designated as E
distance.Formula is as (2):
Wherein, x (M, i), x (M, j) and y (M, i), y (M, j) has represented respectively the transverse and longitudinal coordinate of car light i and car light j.Eps represents that is infinitely close to a positive number of zero, and being used for avoiding denominator here is zero.
2. for the car light existed in same two field picture, calculate the area difference of any two car lights, formula is as (3):
3. for the car light existed in same two field picture, calculate the angle of any two car lights, formula is as (4):
4. calculate the fuzzy membership of each feature of car light according to statistical information, and, according to actual conditions, on feature degree of membership curve, the corresponding degree of membership of first horizontal ordinate is set to 1.Finally calculate total fuzzy membership of any two car lights, two car lights of degree of membership maximum are a pair of, and formula is as (5):
E
M(i,j)=α
1×E
angle+α
2×E
disance+α
3×
darea (5)
Wherein, α
ibe [0,1] interval interior number, by magnitude of traffic flow size, institute is determined.The vehicle flowrate size by pixel in the car light bianry image extracted and size judge.In the high density sequence, α
imeet α
1>α
2>α
3and α
1+ α
2+ α
3=1. other the time α
2>α
1>α
3and α
1+ α
2+ α
3=1.α
icontrolling the importance of each feature in the car light pairing.
5. car light pairing feedback modifiers mechanism: the information of frame car light pairing before and after utilizing, to being revised when front car light pairing situation, can judge the time of blocking generation simultaneously, these shelves are processed, if two car lights of continuous 5 frame are paired with each other, these two car lights belong to a car together.
5), car tracing and feedback modifiers mechanism: according to car light car light pairing situation, vehicle is followed the trail of, introduced car tracing feedback modifiers mechanism simultaneously and guarantee car one track.Concrete grammar is as follows:
1. obtain the situation of car light pairing: bicycle lamp, oneself and oneself pairing; Two car lights, paired with each other; Three car lights, one is own and oneself matches, all the other two are paired with each other; Four car lights are joined team in twos, obtain two pairs of pairing car lights etc.
2. car light is carried out to mark: the center of the car light of each pairing, by a unique ID sign, forms label matrix TG
t=[T
1, T
2..., T
b], T
1, T
2... T
bbe respectively the ID sign of B the lamp cluster centre be detected.
3. follow the trail of vehicle: t+1 is matched to the center of car light line and t constantly and constantly match the center of car light line and couple together, obtain the driving trace of vehicle.
4. car tracing feedback modifiers mechanism: utilize the information such as track slope, track life period, course length, determine whether vehicle driving trace, and eliminate the situation of a car multi-trace.
Several specific question explanations that 2 car tracings run into often:
1. for only having a headlight by the vehicle of other occlusion, and another headlight still can correctly be searched and be followed the trail of, and just can not realize the pairing of two headlights.At this moment the feedback modifiers mechanism of utilizing car light to match can find blocks time of origin, and obtains another car light position that is blocked according to car light spacing feature degree of membership information, and this problem just can obtain solving.
2. for some vehicle, 1,3 or a plurality of headlight are arranged, can utilize equally above-mentioned pairing criterion headlight pairing, just in this link of car tracing, introduce feedback modifiers mechanism, just can remove unnecessary car light track, guarantee car one track;
The accompanying drawing explanation
Fig. 1 vehicle tracing system block diagram
Fig. 2 car light extracts figure
Fig. 3 car light feature degree of membership curve and correction curve
A-c is respectively the statistics degree of membership curve of angle, distance, area features
D-f is respectively the correction degree of membership curve of angle, distance, area features
Fig. 4 vehicle pairing feedback modifiers figure
(PM represents car light pairing matrix, and P (M, j)=i represents the car light i of M frame and car light j pairing.That Flag (M, j)=1 represents continuous 5 frames and car light j pairing is all car light i, judges that car light i and j belong to a car together, otherwise Flag (M, j)=0 judges that car light i and j do not belong to a car).
Fig. 5 car tracing feedback modifiers figure
(TH, DH, SH, CH, th, dh, sh, ch, TT, DD, SS, CC are given threshold value).
The detection of Fig. 6 high density sequence car light and pairing be figure as a result
Fig. 7 avenges day car light detection and the figure (strong reflection) as a result that matches
The detection of Fig. 8 rainy day car light and pairing be figure (motorcycle) as a result
The detection of Fig. 9 bend sequence car light and pairing be figure as a result
The high-speed sequence car light detection of Figure 10 and pairing be figure as a result
Figure 11 car tracing is figure as a result
The a high density
B avenges sky
C rainy day 1 car light and 3 car lights
D rainy days 4 car light
The e bend
F is high-speed
Embodiment
Utilize method for tracing of the present invention to be followed the trail of the vehicle under 6 groups of night complex background conditions.Image is taken under pattern at night by SONY HDR-550D camera, and shooting environmental has considered that high density, high-speed, sleet sky, car light are 1 or for 2, and the situation of bend.Image pixel is as 320*240.The experimental data that it is detected, matches and follows the trail of is added up, and statistics is as listed as table 1-3.
Table 1 vehicle detection result
Table 2 car light pairing result
Table 3 car tracing result
By above data can find out to 6 groups of images be detected as power between 88.13% and 92.37%, be paired into power between 80.00% and 88.89%, follow the trail of accuracy rate up to 92.31%.Illustrate that this paper method detects car light and car tracing has effect preferably.
Experimental result shows: the method for the present invention contrast of image of having utilized homomorphic filter to strengthen, the azimuth ambiguity Techniques For Reducing street lamp, the advertising license plate bright lamp, car light irradiation is the interference to locate lamp area to other luminous objects of the non-car lightings such as the reflected light of other vehicle body and road reflection light, make locate lamp area more accurate, next uses car light feature degree of membership to pairing, add car light pairing feedback modifiers mechanism simultaneously, make the pairing result more accurate, finally utilize the information of pairing car light to be followed the trail of vehicle, and according to car tracing feedback modifiers mechanism, track of vehicle is processed, guarantee car one track, table 1-3 explanation, this method has very strong adaptability and higher accuracy rate for vehicle detection and tracking.
Claims (3)
1. the method that can realize vehicle detection and tracking in complicated traffic video at high efficiency night said method comprising the steps of:
1., figure image intensifying: utilize homomorphic filter, at frequency domain, image is processed, strengthen the contrast of image;
2., car light extracts: to the image after strengthening, application azimuthal blur technology and automatic threshold technology are extracted car light, simultaneously, elimination, because street lamp, billboard, ground return, headlight are got to the interference to locate lamp area such as other vehicle body reflected light, blurring mapping, obtains car light position and shape information accurately;
3., the bicycle lamp follows the trail of: in view of car light motion in image is slowly, and interframe has the area of repetition, so locate to introduce this feature of area severe, follows the trail of car light, and caravan lamp area multiplicity is E
oarea, computing formula is as follows:
Wherein, S
overlap(M, i, M+1, j) represents the area of M frame car light i and M+1 frame car light j repeating part, S (M, i) represents the area of M frame car light i, S (M+1, j) represent the area of M+1 frame car light j, two car lights of area multiplicity maximum, be labeled as a car cresset;
4., the car light pairing is feedback modifiers mechanism: the information of the car light of the successful pairing at first we extracted is added up, calculate again the fuzzy membership of the features such as car (distance is than, area difference, angle) according to statistical information, use the feature Fuzzy degree of membership to be matched to car light, introduce pairing simultaneously and revise feedback mechanism, guarantee the accuracy of pairing, also well solved the marriage problem blocked while occurring;
5., car tracing and feedback modifiers mechanism: according to car light car light pairing situation, vehicle is followed the trail of, introduced car tracing feedback modifiers mechanism simultaneously and guarantee car one track.
2. in night claimed in claim 1 complicated traffic video in the feature of car tracing method, it is as follows that wherein said car light matches concrete grammar used:
1. for the car light existed in same two field picture, the distance of calculating any two car lights than, the ratio of less ordinate in two car light Diff Es and two car lights, be designated as E
distance.Formula is as (2):
Wherein, x (M, i), x (M, j) and y (M, i), y (M, j) has represented respectively the transverse and longitudinal coordinate of car light i and car light j.Eps represents that is infinitely close to a positive number of zero, and being used for avoiding denominator here is zero;
2. for the car light existed in same two field picture, calculate the area difference of any two car lights, formula is as (3):
3. for the car light existed in same two field picture, calculate the angle of any two car lights, formula is as (4):
4. calculate the fuzzy membership of each feature of car light according to statistical information, and, according to actual conditions, on feature degree of membership curve, the corresponding degree of membership of first horizontal ordinate is set to 1;
5. calculate total fuzzy membership of any two car lights, two car lights of degree of membership maximum are a pair of, and formula is as (5):
E
M(i,j)=α
1×E
angle+α
2×E
disance+α
3×E
darea (5)
Wherein, α
ibe [0,1] number in interval, by magnitude of traffic flow size, institute is determined, the vehicle flowrate size by pixel in the car light bianry image extracted and size judge, in the high density sequence, α
imeet α
1>α
2>α
3and α
1+ α
2+ α
3=1, other the time α
2>α
1>α
3and α
1+ α
2+ α
3=1, α
1controlling the importance of each feature in the car light pairing.
3. in night claimed in claim 1 complicated traffic video in the feature of car tracing method, wherein said car tracing concrete grammar used is as follows:
1. obtain the situation of car light pairing: the bicycle lamp, oneself and oneself pairing, two car lights, paired with each other, three car lights, one is that oneself matches with oneself, and all the other two are paired with each other, and four car lights are joined team in twos, obtain two pairs of pairing car lights etc.;
2. car light is carried out to mark: the center of the car light of each pairing, by a unique ID sign, forms label matrix TG
t=[T
1, T
2..., T
b], T
1, T
2... T
bbe respectively the ID sign of B the lamp cluster centre be detected;
3. follow the trail of vehicle: t+1 is matched to the center of car light line and t constantly and constantly match the center of car light line and couple together, obtain the driving trace of vehicle.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013104066535A CN103440771A (en) | 2013-09-06 | 2013-09-06 | Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013104066535A CN103440771A (en) | 2013-09-06 | 2013-09-06 | Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103440771A true CN103440771A (en) | 2013-12-11 |
Family
ID=49694464
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013104066535A Pending CN103440771A (en) | 2013-09-06 | 2013-09-06 | Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103440771A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106406526A (en) * | 2016-09-07 | 2017-02-15 | 长安大学 | Auxiliary car lamp control method capable of pre-judging steering intention of driver |
CN106408938A (en) * | 2016-09-13 | 2017-02-15 | 天津工业大学 | Complete extraction method of various vehicle tracks in urban traffic monitoring at night |
CN107038711A (en) * | 2016-02-02 | 2017-08-11 | 财团法人资讯工业策进会 | Adaptive evolution type vehicle lamp signal detection tracking and identification system and method |
CN107292260A (en) * | 2017-06-15 | 2017-10-24 | 武汉理工大学 | The thick fog day vehicle checking method of pairing is associated with fog lamp based on vehicle head lamp |
CN107506686A (en) * | 2017-07-12 | 2017-12-22 | 浙江工业大学 | A kind of vehicle detection at night method based on conspicuousness detection |
CN108538052A (en) * | 2018-03-05 | 2018-09-14 | 华南理工大学 | Night traffic flow rate testing methods based on headlight track following and dynamic pairing |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7577274B2 (en) * | 2003-09-12 | 2009-08-18 | Honeywell International Inc. | System and method for counting cars at night |
CN103050008A (en) * | 2013-01-14 | 2013-04-17 | 哈尔滨工程大学 | Method for detecting vehicles in night complex traffic videos |
-
2013
- 2013-09-06 CN CN2013104066535A patent/CN103440771A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7577274B2 (en) * | 2003-09-12 | 2009-08-18 | Honeywell International Inc. | System and method for counting cars at night |
CN103050008A (en) * | 2013-01-14 | 2013-04-17 | 哈尔滨工程大学 | Method for detecting vehicles in night complex traffic videos |
Non-Patent Citations (2)
Title |
---|
WEI ZHANG ET AL.: "Tracking and Pairing Vehicle Headlight in Night Scenes", 《INTELLIGENT TRANSPORTATION SYSTEMS》 * |
刘莉等: "基于视频的夜间车流量统计", 《微处理机》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107038711A (en) * | 2016-02-02 | 2017-08-11 | 财团法人资讯工业策进会 | Adaptive evolution type vehicle lamp signal detection tracking and identification system and method |
CN106406526A (en) * | 2016-09-07 | 2017-02-15 | 长安大学 | Auxiliary car lamp control method capable of pre-judging steering intention of driver |
CN106406526B (en) * | 2016-09-07 | 2019-07-26 | 长安大学 | A kind of auxiliary vehicle light control method that can be prejudged driver and turn to intention |
CN106408938A (en) * | 2016-09-13 | 2017-02-15 | 天津工业大学 | Complete extraction method of various vehicle tracks in urban traffic monitoring at night |
CN106408938B (en) * | 2016-09-13 | 2019-09-06 | 天津工业大学 | The complete extraction of various types of vehicles track in the monitoring of night urban transportation |
CN107292260A (en) * | 2017-06-15 | 2017-10-24 | 武汉理工大学 | The thick fog day vehicle checking method of pairing is associated with fog lamp based on vehicle head lamp |
CN107506686A (en) * | 2017-07-12 | 2017-12-22 | 浙江工业大学 | A kind of vehicle detection at night method based on conspicuousness detection |
CN108538052A (en) * | 2018-03-05 | 2018-09-14 | 华南理工大学 | Night traffic flow rate testing methods based on headlight track following and dynamic pairing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10262216B2 (en) | Hazard detection from a camera in a scene with moving shadows | |
Varma et al. | IDD: A dataset for exploring problems of autonomous navigation in unconstrained environments | |
EP2549457B1 (en) | Vehicle-mounting vehicle-surroundings recognition apparatus and vehicle-mounting vehicle-surroundings recognition system | |
CN107031656B (en) | Virtual sensor data generation for wheel immobilizer detection | |
Wang et al. | Performance and challenges of 3D object detection methods in complex scenes for autonomous driving | |
CN103440771A (en) | Application of fuzzy membership grade and feedback correction in night traffic video vehicle detection | |
Che et al. | D $^ 2$-city: a large-scale dashcam video dataset of diverse traffic scenarios | |
Yu et al. | Traffic light detection during day and night conditions by a camera | |
Borkar et al. | A layered approach to robust lane detection at night | |
CN103050008B (en) | Method for detecting vehicles in night complex traffic videos | |
CN106934374B (en) | Method and system for identifying traffic signboard in haze scene | |
Lin et al. | Lane departure and front collision warning using a single camera | |
CN113052159B (en) | Image recognition method, device, equipment and computer storage medium | |
Chiu et al. | Automatic Traffic Surveillance System for Vision-Based Vehicle Recognition and Tracking. | |
CN103129468A (en) | Vehicle-mounted roadblock recognition system and method based on laser imaging technique | |
CN109886175A (en) | A kind of method for detecting lane lines that straight line is combined with circular arc | |
KR101134857B1 (en) | Apparatus and method for detecting a navigation vehicle in day and night according to luminous state | |
CN105740835A (en) | Preceding vehicle detection method based on vehicle-mounted camera under night-vision environment | |
Reyad et al. | Traffic conflict–based before–after study with use of comparison groups and the empirical Bayes method | |
CN104658249A (en) | Method for rapidly detecting vehicle based on frame difference and light stream | |
CN110852157A (en) | Deep learning track line detection method based on binarization network | |
Ku et al. | Visual motorcycle detection and tracking algorithms | |
Boonsim et al. | An algorithm for accurate taillight detection at night | |
Deng et al. | An integrated forward collision warning system based on monocular vision | |
CN103886292A (en) | Night vehicle target stable tracking method based on machine vision |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131211 |