CN104992160A - Night preceding vehicle detection method for heavy-duty truck - Google Patents
Night preceding vehicle detection method for heavy-duty truck Download PDFInfo
- Publication number
- CN104992160A CN104992160A CN201510419932.4A CN201510419932A CN104992160A CN 104992160 A CN104992160 A CN 104992160A CN 201510419932 A CN201510419932 A CN 201510419932A CN 104992160 A CN104992160 A CN 104992160A
- Authority
- CN
- China
- Prior art keywords
- value
- image
- gray level
- classifier
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Traffic Control Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a night preceding vehicle detection method for heavy-duty trucks. The night preceding vehicle detection method comprises the steps of acquiring a classifier and realizing vehicle detection. Specifically, the classifier acquisition comprises the steps of: removing interference in a gray scale image of a driving environment in front of a heavy-duty truck by adopting a threshold value processing method; intercepting a vehicle-lamp-pair region as a positive sample, and intercepting non-vehicle-lamp-pair regions as negative samples; and training the positive sample and the negative samples by adopting an adaboost algorithm based on haar-like features to obtain the classifier. The vehicle detection realization comprises the steps of reading a current frame gray scale image of video in real time and executing the operations as following: removing the interference in the current frame gray scale image by adopting the threshold value processing method to obtain a detected and processed current frame gray scale image; loading the classifier; detecting a vehicle-lamp-pair region in the detected and processed current frame gray scale image; and marking the vehicle-lamp-pair region in a copy of the current frame gray scale image. The night preceding vehicle detection method for heavy-duty trucks removes tail lamp interference, preserves shape of the vehicle lamp pair perfectly, reduces interference, simplifies the number of samples, improves the detection rate of the classifier, marks the detection result in the original image, and verifies the practicability of the device.
Description
Technical field
The present invention relates to a kind of heavy truck front vehicles at night detection method, belong to technical field of vehicle safety.
Background technology
The fast development of the communication and logistics industry that the advance of economic society brings brings very large convenience to the life of society and people, and meanwhile, automobile quantity sharply increases, and which results in going from bad to worse of road traffic condition.DAS (Driver Assistant System) (DAS) can help driver's operating and controlling vehicle by the running environment analyzing Current vehicle, and improve traffic safety, Accident prevention occurs.The input of DAS (Driver Assistant System) is generally digital picture (as from CCD camera or CMOS camera), infrared view, laser, radar, ultrasound wave or gps signal.
The traditional night vision technology come into operation mainly contains three major types: Low-light Level Night Vision Technology, active infrared night-viewing technology, infrared thermal imaging technique.Faint natural light image is changed into by image amplifier and enhances hundred times even electronic image of several ten thousand times by low-light night vision system, then the electronic image of enhancing is transformed into visual optical imagery.Low-light level night vision device can only have moonlight or starlight and smoglessly work at night, and operating distance declines with brightness decline rate.Militarily, main as helmet night vision eyes, use mainly as sighting telescope for light arms, nighttime driving and suggestion observation care unit at present.Active infrared night-viewing system is made up of infrared searchlight and infrared viewer.Infrared searchlight for launching the sightless wavelength coverage of the human eye i.e. near infrared light of 0.9 ~ 1.2 micron, to irradiate observed scenery; Infrared viewer is that the infrared image that observed reflected light is formed is transformed into human viewable's image.The operating distance of active infrared night vision sight depends on the power of worn infrared searchlight, and power is larger, and operating distance is far away.Active infrared night vision sight is mainly used in the military equipment such as rifle, tank.Also not needing the light source added during the work of infrared thermal imaging night vision system, is a kind of passive night vision system.But, the principle of work of it and low-light level night vision device is completely different, low-light level night vision device relies on enhancing natural light to carry out work, and thermal infrared imager utilizes the temperature difference at each position of scenery itself and the temperature difference between scenery and background thereof to carry out imaging, namely infra-red thermal imaging system is that the infrared light self sent by direct receiving target object carrys out imaging, has very high verification and measurement ratio.Night vision technology all has very high detection performance, but price costly, generally all for national defense and military department of various countries.Just think, on each automobile, above-mentioned any one night vision device is installed, all need certain cost.And these traditional night vision instruments all have stronger specific aim, with regard to a kind of night vision device, can not all show good effect in each environment.
Affect the object of Current vehicle mainly for detection of the pedestrian around Current vehicle, barrier, traffic sign and other vehicles etc. based on the DAS (Driver Assistant System) of computer vision, the detection of front vehicles effectively can reduce the generation of traffic hazard.Vehicle detecting system at night based on computer vision is more suitable for general-utility car and uses, the algorithm development relative maturity of each side such as present computer vision, image procossing and pattern-recognition, computer hardware performance also increases substantially, and use common CCD camera just can gather required video data, this research just making employing computer vision methods carry out vehicle early warning at night has all had guarantee from cost, algorithm, hardware or software aspect.Carry out that vehicle detection illumination is strong, visibility is high daytime, the characteristic informations such as color of object, edge, profile, symmetry are given prominence to, and are easy to extract vehicle target information, vehicle detecting algorithm relative maturity on daytime track.Nighttime driving scene illumination condition is poor, and driver's sighting distance is not good, and nighttime driving becomes worrying, one of the driving condition of Frequent Accidents.
Because illumination is not enough, for daytime vehicle detection method and be not suitable for night.Under nighttime conditions, the taillight of a pair symmetry is the most obvious feature that front vehicles presents, and is also our unique utilizable Detection Information.Some researchers utilize color and the movable information positioned vehicle of taillight, by analyzing taillight region R, G, B tri-channel informations, appropriate threshold is set, extract the pixel that red intensity value is greater than threshold value, then judge whether two car lights belong to same car by the spatial relationship of checking two taillights, and then determine vehicle location.In addition, many researchers analyze the symmetry of a pair car light of same car, judge whether to belong to same vehicle by the information such as size, area contrasting each high-brightness region block.At present, along with the fast development of integrated chip, have researcher to propose to use Haar-like character representation front vehicles region, use Adaboost algorithm training vehicle detection at night sorter, the method is than extracting feature merely, having robustness according to taillight symmetry positioned vehicle.
The method of conventional exercises sorter is from original image, intercept positive negative sample, and detect in image after former figure or smoothing processing, such training classifier needs more positive negative sample, and because multiple disturbing factor exists very large flase drop in testing process.
But, when heavy truck travels on expressway, often travel speed is fast, and truck headlight is very bright, road surface also exists the interference such as reflective of the object such as the bright territory of sheet, road sign, road surface, guardrail, label formed due to the highlighted headlight of heavy truck, also make the light for vehicle of the different lanes in front present small one and large one, obvious asymmetric situation, this considerably increases the difficulty detecting heavily card front vehicles night, make said method can not the interference in good other clear zones of filtering, follow-up testing is impacted.
Summary of the invention
Not enough for prior art, the invention discloses a kind of heavy truck front vehicles at night detection method;
The present invention travels the gray level image pre-service of traveling ahead environment to the heavy truck of shooting, remove other interfere informations as far as possible, such as, the large area clear zone formed on road surface by the headlight that heavy truck is highlighted and guardrail, the reflective interference formed of car plate, and the shape that each class light for vehicle of intact reservation front is right.To pretreated image interception light for vehicle to part, light for vehicle is to the positive sample of part as training classifier, and other region of pretreated image is as the negative sample of training classifier.Use and based on the adaboost algorithm of haar-like feature, the positive and negative samples of training classifier is trained, finally detected front vehicles sorter at night, use in this sorter image after the pre-treatment and carry out vehicle detection, and mark vehicle location in original image.Owing to only extracting the right shape of light for vehicle in pretreated image, the positive sample mode of training classifier reduces, and the negative sample of training classifier also simplifies thereupon, and detects in image after the pre-treatment, testing environment simplifies greatly, has higher accuracy and stronger robustness.
Technical scheme of the present invention is:
A kind of heavy truck front vehicles at night detection method, concrete steps comprise:
A, acquisition sorter
(1) in heavy truck driving process, shooting heavy truck travels traveling ahead environment, obtains 8 a large amount of gray level images;
(2) interference in each the frame gray level image in 8 a large amount of gray level images adopting thresholding method removal step (1) to obtain;
(3), in the gray level image obtained in step (2), intercept car light to the positive sample of region as training classifier, intercept non-car light to the negative sample of region as training classifier;
(4) use the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtain sorter;
B, realize vehicle detection
(5) read the present frame gray image of video in real time and perform following operation: copy present frame gray image, employing thresholding method removes the interference in described present frame gray image, obtain the present frame gray image after check processing, the sorter that load step (4) obtains, detect car light in the present frame gray image after described check processing to region, export car light to region, and get the bid caravan lamp to region at the present frame gray image of described copy.
Described video refers to that the heavy truck of captured in real-time in heavy truck driving process travels the video of traveling ahead environment.
Preferred according to the present invention, described thresholding method, specifically comprises:
A, calculating gray level image are at area-of-interest (0, N
cols/ m, N
rows, N
cols(m-1)/m) pixel value average μ
l, described area-of-interest refers to: the position of first pixel in the upper left corner is (0, N
cols/ m), wide is N
rows, height is N
cols(m-1)/m; The span of m is that the span of 2 ~ 5, m makes area-of-interest not comprise sky portion.μ
lcomputing formula such as formula shown in (I):
In formula (I), N
colsrepresent the number of pixels often arranged of gray level image, N
rowsrepresent the number of pixels of often going of gray level image, N
cols× N
rowsrefer to the size of gray level image, I
t(x, y) refers to the pixel value that in gray level image, (x, y) puts;
B, on gray level image, delimit area-of-interest (0, k*N
cols/ N, N
rows, N
cols/ N), 1≤k≤N, N represents the number of the area-of-interest of delimiting at equal intervals successively on gray level image, and k initial value is 1;
C, in a kth area-of-interest, setting size be the wicket of m × n, m < < N
cols, n < < N
rows; Make i=1;
A wicket traversal kth area-of-interest described in d, use step c: if the local pixel average u in described wicket
ibe greater than u
l, and the local maximum gradation value M in described wicket
ibe greater than M, then calculate the Local standard deviation σ in described wicket
i, calculate the local threshold T in described wicket
i=M
i-σ
i, i adds 1, and wherein, the span of M is 240-254;
E, try to achieve i threshold value, i.e. T at a kth area-of-interest
k1, T
k2... T
kiif k < N, gets
as the threshold value of a kth area-of-interest, k adds 1, enters step c; If k=N, get T
tN=max{T
k1... T
ki, enter step f;
N number of area-of-interest of f, gray level image tries to achieve N number of different threshold value T
t1, T
t2..., T
tN.
Preferred according to the present invention, calculate the Local standard deviation σ in described wicket
i, specific formula for calculation is such as formula shown in (II):
In formula (II), Ω represents gray level image.
Preferred according to the present invention, use the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtain sorter; Concrete steps comprise:
G, use positive sample and negative sample described in the adaboost Algorithm for Training based on haar-like feature, setting training parameter, described training parameter comprises: the progression nstage of training classifier, determine the sorter nsplits being used for Stage Classification device, for training the positive sample number npos of every first-level class device, for training the negative sample number nneg of every first-level class device, the minimum hit rate minhitrate that every first-level class device needs, the maximum false alarm rate maxfalsealarm of every first-level class device, the wide w of positive sample or negative sample, the high h of positive sample or negative sample, free memory mem, select Adaboost algorithm type bt, the span of nstage is 12-20, nsplits is 1 or 2, the span of npos is 500-1205, the span of nneg is 1500-2540, the span of minhitrate is 0.95--0.999, the value of maxfalsealarm is 0.5, the value of w is 20, the value of h is 20, mem value is 512MB, bt is discrete Adaboost algorithm DAB, obtain every layer of optimum Weak Classifier and given weights, by some Weak Classifier composition strong classifiers, multistage strong classifier forms cascade classifier, i.e. step (4) described sorter.
The executable file opencv_haartraining.exe called in existing OpenCv realizes the adaboost algorithm based on haar-like feature, and training parameter, as input value, can realize operating described in step g.
Experimentally verify, the sorter that the setting of above-mentioned training parameter obtains more is conducive to the detection of heavy truck front vehicles at night.
According to the present invention preferably, the value of nstage is the value of 14, nsplits is 1, represent simple two classification sorters, the value of npos is the span of 900, nneg is 1900, the value of minhitrate is 0.998, bt is discrete Adaboost algorithm DAB.
Preferred according to the present invention, the value of M is 250;
Beneficial effect of the present invention is:
1, thresholding method of the present invention has removed car light interference largely non-, and intactly remain the right shape of car light, to originally intercept the work simplification of a large amount of positive sample for intercepting all kinds of car light to shape, interference reduces, negative sample quantity reduces, enormously simplify sample size, shorten the training time, improve the verification and measurement ratio of sorter; Interfere information in image after threshold process, almost completely by filtering, uses in sorter image after treatment and detects, reduce and have detected the time, enhance robustness; In former figure, mark testing result, demonstrate the practicality of device;
2, the present invention greatly reduces the threshold process time by the design conditions limiting each regional area, and effective filtering clear zone, road surface, local threshold is tried to achieve at qualified regional area, choose the threshold value of maximum local threshold as entire image, this method effectively eliminates interfere information, particularly owing to heavily blocking large stretch of clear zone that highlighted headlight is formed on road surface, and the shape that intact reservation car light is right, make car light more obvious to place Haar-like feature;
3, the adaptive multi-thresholding method for solving improved and threshold process being combined with sorter, prevention traffic hazard tired to mitigation heavy truck driver nighttime driving is very helpful, and has very large researching value and use meaning.
Accompanying drawing explanation
Fig. 1 is workflow diagram of the present invention;
Fig. 2 is the piece image schematic diagram in a large amount of 8 gray level images of the present invention;
Fig. 3 adopts thresholding method to remove the image schematic diagram that the interference in Fig. 2 obtains afterwards;
Fig. 4 is present frame gray image schematic diagram in embodiment 1;
Fig. 5 is the present frame gray image schematic diagram in embodiment 1 after check processing;
Fig. 6 is the present frame gray image schematic diagram in embodiment 1 after mark.
Embodiment
Below in conjunction with Figure of description and embodiment, the present invention is further qualified, but is not limited thereto.
Embodiment 1
A kind of heavy truck front vehicles at night detection method, concrete steps comprise:
A, acquisition sorter
(1) in heavy truck driving process, shooting heavy truck travels traveling ahead environment, obtains 8 a large amount of gray level images; Fig. 2 is the piece image schematic diagram in described a large amount of 8 gray level images;
(2) interference in each the frame gray level image in 8 a large amount of gray level images adopting thresholding method removal step (1) to obtain; The image schematic diagram that interference in removal Fig. 2 obtains afterwards as shown in Figure 3;
(3), in the gray level image obtained in step (2), intercept car light to the positive sample of region as training classifier, intercept non-car light to the negative sample of region as training classifier;
(4) use the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtain sorter;
B, realize vehicle detection
(5) read the present frame gray image of video in real time and perform following operation: present frame gray image I
tas shown in Figure 4, copy present frame gray image, employing adaptive thresholding method removes the interference in described present frame gray image, obtain the present frame gray image after check processing, as shown in Figure 5, the sorter that load step (4) obtains, detects car light in the present frame gray image after described check processing to region, export car light to region, and get the bid caravan lamp to region at the present frame gray image of described copy.As shown in Figure 6.Described video refers to that the heavy truck of captured in real-time in heavy truck driving process travels the video of traveling ahead environment.
Embodiment 2
Heavy truck front vehicles at night detection method according to embodiment 1, its difference is, described thresholding method, specifically comprises:
A, calculating gray level image are at area-of-interest (0, N
cols/ m, N
rows, N
cols(m-1)/m) pixel value average μ
l, described area-of-interest refers to: the position of first pixel in the upper left corner is (0, N
cols/ m), wide is N
rows, height is N
cols(m-1)/m; The value of m is that the span of 3, m makes area-of-interest not comprise sky portion.μ
lcomputing formula such as formula shown in (I):
In formula (I), N
colsrepresent the number of pixels often arranged of gray level image, N
rowsrepresent the number of pixels of often going of gray level image, N
cols× N
rowsrefer to the size of gray level image, I
t(x, y) refers to the pixel value that in gray level image, (x, y) puts;
B, on gray level image, delimit area-of-interest (0, k*N
cols/ N, N
rows, N
cols/ N), 1≤k≤N, N represents the number of the area-of-interest of delimiting at equal intervals successively on gray level image, and k initial value is 1;
C, in a kth area-of-interest, setting size be the wicket of m × n, m < < N
cols, n < < N
rows; Make i=1;
A wicket traversal kth area-of-interest described in d, use step c: if the local pixel average u in described wicket
ibe greater than u
l, and the local maximum gradation value M in described wicket
ibe greater than M, then calculate the Local standard deviation σ in described wicket
i, calculate the local threshold T in described wicket
i=M
i-σ
i, i adds 1, and wherein, the value of M is 250;
E, try to achieve i threshold value, i.e. T at a kth area-of-interest
k1, T
k2... T
kiif k < N, gets
as the threshold value of a kth area-of-interest, k adds 1, enters step c; If k=N, get T
tN=max{T
k1... T
ki, enter step f;
N number of area-of-interest of f, gray level image tries to achieve N number of different threshold value T
t1, T
t2..., T
tN.
Embodiment 3
Heavy truck front vehicles at night detection method according to embodiment 2, its difference is, calculates the Local standard deviation σ in described wicket
i, specific formula for calculation is such as formula shown in (II):
In formula (II), Ω represents gray level image.
Embodiment 4
Heavy truck front vehicles at night detection method according to embodiment 1, its difference is, uses the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtains sorter; Concrete steps comprise:
G, use positive sample and negative sample described in the adaboost Algorithm for Training based on haar-like feature, setting training parameter, described training parameter comprises: the progression nstage of training classifier, determine the sorter nsplits being used for Stage Classification device, for training the positive sample number npos of every first-level class device, for training the negative sample number nneg of every first-level class device, the minimum hit rate minhitrate that every first-level class device needs, the maximum false alarm rate maxfalsealarm of every first-level class device, the wide w of positive sample or negative sample, the high h of positive sample or negative sample, free memory mem, select Adaboost algorithm type bt, the value of nstage is 14, nsplits is 1, represent simple two classification sorters, the value of npos is 900, the value of nneg is 1900, the value of minhitrate is 0.998, the value of maxfalsealarm is 0.5, the value of w is 20, the value of h is 20, mem value is 512MB, bt is discrete Adaboost algorithm DAB, obtain every layer of optimum Weak Classifier and given weights, by some Weak Classifier composition strong classifiers, multistage strong classifier forms cascade classifier, i.e. step (4) described sorter.
The executable file opencv_haartraining.exe called in existing OpenCv realizes the adaboost algorithm based on haar-like feature, and training parameter, as input value, can realize operating described in step g.
Experimentally verify, the sorter that the setting of above-mentioned training parameter obtains more is conducive to the detection of heavy truck front vehicles at night.
Claims (6)
1. heavy truck front vehicles at a night detection method, it is characterized in that, concrete steps comprise:
A, acquisition sorter
(1) in heavy truck driving process, shooting heavy truck travels traveling ahead environment, obtains 8 a large amount of gray level images;
(2) interference in each the frame gray level image in 8 a large amount of gray level images adopting thresholding method removal step (1) to obtain;
(3), in the gray level image obtained in step (2), intercept car light to the positive sample of region as training classifier, intercept non-car light to the negative sample of region as training classifier;
(4) use the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtain sorter;
B, realize vehicle detection
(5) read the present frame gray image of video in real time and perform following operation: copy present frame gray image, employing thresholding method removes the interference in described present frame gray image, obtain the present frame gray image after check processing, the sorter that load step (4) obtains, detect car light in the present frame gray image after described check processing to region, export car light to region, and get the bid caravan lamp to region at the present frame gray image of described copy.
2. heavy truck front vehicles at night detection method according to claim 1, it is characterized in that, described thresholding method, specifically comprises:
A, calculating gray level image are at area-of-interest (0, N
cols/ m, N
rows, N
cols(m-1)/m) pixel value average μ
l, described area-of-interest refers to: the position of first pixel in the upper left corner is (0, N
cols/ m), wide is N
rows, height is N
cols(m-1)/m; The span of m is 2 ~ 5, μ
lcomputing formula such as formula shown in (I):
In formula (I), N
colsrepresent the number of pixels often arranged of gray level image, N
rowsrepresent the number of pixels of often going of gray level image, N
cols× N
rowsrefer to the size of gray level image, I
t(x, y) refers to the pixel value that in gray level image, (x, y) puts;
B, on gray level image, delimit area-of-interest (0, k*N
cols/ N, N
rows, N
cols/ N), 1≤k≤N, N represents the number of the area-of-interest of delimiting at equal intervals successively on gray level image, and k initial value is 1;
C, in a kth area-of-interest, setting size be the wicket of m × n, m < < N
cols, n < < N
rows; Make i=1;
A wicket traversal kth area-of-interest described in d, use step c: if the local pixel average u in described wicket
ibe greater than u
l, and the local maximum gradation value M in described wicket
ibe greater than M, then calculate the Local standard deviation σ in described wicket
i, calculate the local threshold T in described wicket
i=M
i-σ
i, i adds 1, and wherein, the span of M is 240-254;
E, try to achieve i threshold value, i.e. T at a kth area-of-interest
k1, T
k2... T
kiif k < N, gets
as the threshold value of a kth area-of-interest, k adds 1, enters step c; If k=N, get T
tN=max{T
k1... T
ki, enter step f;
N number of area-of-interest of f, gray level image tries to achieve N number of different threshold value T
t1, T
t2..., T
tN.
3. heavy truck front vehicles at night detection method according to claim 2, is characterized in that, calculate the Local standard deviation σ in described wicket
i, specific formula for calculation is such as formula shown in (II):
In formula (II), Ω represents gray level image.
4. heavy truck front vehicles at night detection method according to claim 1, is characterized in that, use the described positive sample of adaboost Algorithm for Training step (3) based on haar-like feature and negative sample, obtain sorter; Concrete steps comprise:
G, use positive sample and negative sample described in the adaboost Algorithm for Training based on haar-like feature, setting training parameter, described training parameter comprises: the progression nstage of training classifier, determine the sorter nsplits being used for Stage Classification device, for training the positive sample number npos of every first-level class device, for training the negative sample number nneg of every first-level class device, the minimum hit rate minhitrate that every first-level class device needs, the maximum false alarm rate maxfalsealarm of every first-level class device, the wide w of positive sample or negative sample, the high h of positive sample or negative sample, free memory mem, select Adaboost algorithm type bt, the span of nstage is 12-20, nsplits is 1 or 2, the span of npos is 500-1205, the span of nneg is 1500-2540, the span of minhitrate is 0.95--0.999, the value of maxfalsealarm is 0.5, the value of w is 20, the value of h is 20, mem value is 512MB, bt is discrete Adaboost algorithm DAB, obtain every layer of optimum Weak Classifier and given weights, by some Weak Classifier composition strong classifiers, multistage strong classifier forms cascade classifier, i.e. step (4) described sorter.
5. heavy truck front vehicles at night detection method according to claim 4, it is characterized in that, the value of nstage is 14, the value of nsplits is 1, represent simple two classification sorters, the value of npos is the span of 900, nneg is 1900, the value of minhitrate is 0.998, bt is discrete Adaboost algorithm DAB.
6. heavy truck front vehicles at night detection method according to claim 2, it is characterized in that, the value of M is 250.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510419932.4A CN104992160B (en) | 2015-07-16 | 2015-07-16 | A kind of heavy truck night front vehicles detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510419932.4A CN104992160B (en) | 2015-07-16 | 2015-07-16 | A kind of heavy truck night front vehicles detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104992160A true CN104992160A (en) | 2015-10-21 |
| CN104992160B CN104992160B (en) | 2018-08-03 |
Family
ID=54303973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510419932.4A Active CN104992160B (en) | 2015-07-16 | 2015-07-16 | A kind of heavy truck night front vehicles detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104992160B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389547A (en) * | 2015-10-22 | 2016-03-09 | 四川膨旭科技有限公司 | System for identifying vehicle at night |
| CN107316002A (en) * | 2017-06-02 | 2017-11-03 | 武汉理工大学 | A kind of night front vehicles recognition methods based on Active Learning |
| CN110188693A (en) * | 2019-05-30 | 2019-08-30 | 重庆大学 | Improved complex environment vehicle characteristics extract and parking method of discrimination |
| CN112440991A (en) * | 2020-12-02 | 2021-03-05 | 北奔重型汽车集团有限公司 | Auxiliary early warning device applied to heavy truck |
| CN116739457A (en) * | 2023-08-11 | 2023-09-12 | 北京博数智源人工智能科技有限公司 | Production state data processing method and system for surface mine safety analysis |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010006361A1 (en) * | 2008-07-14 | 2010-01-21 | National Ict Australia Limited | Detection of vehicles in images of a night time scene |
| CN103034843A (en) * | 2012-12-07 | 2013-04-10 | 电子科技大学 | Method for detecting vehicle at night based on monocular vision |
| CN103942560A (en) * | 2014-01-24 | 2014-07-23 | 北京理工大学 | High-resolution video vehicle detection method in intelligent traffic monitoring system |
-
2015
- 2015-07-16 CN CN201510419932.4A patent/CN104992160B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010006361A1 (en) * | 2008-07-14 | 2010-01-21 | National Ict Australia Limited | Detection of vehicles in images of a night time scene |
| CN103034843A (en) * | 2012-12-07 | 2013-04-10 | 电子科技大学 | Method for detecting vehicle at night based on monocular vision |
| CN103942560A (en) * | 2014-01-24 | 2014-07-23 | 北京理工大学 | High-resolution video vehicle detection method in intelligent traffic monitoring system |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389547A (en) * | 2015-10-22 | 2016-03-09 | 四川膨旭科技有限公司 | System for identifying vehicle at night |
| CN107316002A (en) * | 2017-06-02 | 2017-11-03 | 武汉理工大学 | A kind of night front vehicles recognition methods based on Active Learning |
| CN110188693A (en) * | 2019-05-30 | 2019-08-30 | 重庆大学 | Improved complex environment vehicle characteristics extract and parking method of discrimination |
| CN110188693B (en) * | 2019-05-30 | 2023-04-07 | 重庆大学 | Improved complex environment vehicle feature extraction and parking discrimination method |
| CN112440991A (en) * | 2020-12-02 | 2021-03-05 | 北奔重型汽车集团有限公司 | Auxiliary early warning device applied to heavy truck |
| CN116739457A (en) * | 2023-08-11 | 2023-09-12 | 北京博数智源人工智能科技有限公司 | Production state data processing method and system for surface mine safety analysis |
| CN116739457B (en) * | 2023-08-11 | 2023-10-20 | 北京博数智源人工智能科技有限公司 | Production state data processing method and system for surface mine safety analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104992160B (en) | 2018-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kurihata et al. | Rainy weather recognition from in-vehicle camera images for driver assistance | |
| CN106169244B (en) | Device and method is provided using the guidance information of crossing recognition result | |
| CN105922991B (en) | Based on the lane departure warning method and system for generating virtual lane line | |
| Chen et al. | Nighttime brake-light detection by Nakagami imaging | |
| Wu et al. | Applying a functional neurofuzzy network to real-time lane detection and front-vehicle distance measurement | |
| CN104992160A (en) | Night preceding vehicle detection method for heavy-duty truck | |
| CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
| CN101670835A (en) | Road-lane marker detection | |
| CN107886034A (en) | Driving based reminding method, device and vehicle | |
| Lin et al. | Adaptive IPM-based lane filtering for night forward vehicle detection | |
| CN103927548A (en) | Novel vehicle collision avoiding brake behavior detection method | |
| CN104268859A (en) | Image preprocessing method for night lane line detection | |
| Chen | Road vehicle recognition algorithm in safety assistant driving based on artificial intelligence | |
| Wen et al. | A rear-vehicle detection system for static images based on monocular vision | |
| CN104463170A (en) | Unlicensed vehicle detecting method based on multiple detection under gate system | |
| CN105787456A (en) | Method for detecting pedestrians in night far infrared images | |
| Nguwi et al. | Number plate recognition in noisy image | |
| Boumediene et al. | Vehicle detection algorithm based on horizontal/vertical edges | |
| US8971579B2 (en) | Windshield localization for occupancy detection | |
| Nagumo et al. | Extraction of forward vehicles by front-mounted camera using brightness information | |
| Álvarez et al. | Perception advances in outdoor vehicle detection for automatic cruise control | |
| Pillai et al. | Night time vehicle detection using tail lights: a survey | |
| Chen et al. | Robust rear light status recognition using symmetrical surfs | |
| Huang et al. | Nighttime vehicle detection and tracking base on spatiotemporal analysis using RCCC sensor | |
| Kumar | Vehicle detection in monocular night-time grey-level videos |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200107 Address after: 250002 room 1010, 10 / F, building a1-4, Hanyu Jinggu, jingshidong Road, Jinan area, China (Shandong) pilot Free Trade Zone, Jinan City, Shandong Province Patentee after: Shandong zhikan Shenjian Information Technology Co., Ltd Address before: Licheng Alexander Road in Ji'nan City, Shandong province 250199 No. 27 Patentee before: Shandong University |