CN108319906A - Pedestrian detection method based on vehicle mounted infrared video and system - Google Patents
Pedestrian detection method based on vehicle mounted infrared video and system Download PDFInfo
- Publication number
- CN108319906A CN108319906A CN201810073180.4A CN201810073180A CN108319906A CN 108319906 A CN108319906 A CN 108319906A CN 201810073180 A CN201810073180 A CN 201810073180A CN 108319906 A CN108319906 A CN 108319906A
- Authority
- CN
- China
- Prior art keywords
- track
- pedestrian
- detection
- image
- window
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 188
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 25
- 230000003044 adaptive effect Effects 0.000 claims abstract description 16
- 230000002708 enhancing effect Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 37
- 239000011159 matrix material Substances 0.000 claims description 36
- 238000012216 screening Methods 0.000 claims description 16
- 239000000284 extract Substances 0.000 claims description 10
- 238000007689 inspection Methods 0.000 claims description 9
- 238000012417 linear regression Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000005855 radiation Effects 0.000 description 6
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 239000002537 cosmetic Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 239000003595 mist Substances 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (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)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of pedestrian detection method and system based on vehicle mounted infrared video, is related to the computer vision field of pedestrian detection.This approach includes the following steps：Infrared video is obtained by vehicular infrared detection instrument, and the color characteristic and Gradient Features of infrared video are extracted using integrogram and channel characteristics；Using adaptive enhancing iterative algorithm Adaboost, the feature of extraction is trained and is screened, the pedestrian in the image of infrared video is detected, obtains Preliminary detection result；Using Hungary Algorithm, Preliminary detection result is calculated, the pedestrian in all images is obtained and is associated with the initial track sequence to be formed；Optimal Track association information is calculated, final pursuit path is obtained using Man Kelaisi algorithms using initial track sequence as initial value.The present invention will be detected and is combined with tracking, can be detected the pedestrian being blocked in the process of moving, be promoted the accuracy rate of pedestrian detection.
Description
Technical field
The present invention relates to the computer vision fields of pedestrian detection, are specifically related to a kind of row based on vehicle mounted infrared video
People's detection method and system.
Background technology
According to the classification of detection, pedestrian detection can be divided into method based on kinetic characteristic, based on template matches
Method and method based on statistical learning.Compared to first two pedestrian detection method, the method based on statistical learning has inspection
It is high to survey precision, robustness is the emphasis and hot spot of current pedestrian detection research preferably a little.Method based on statistical learning is logical
A large amount of pedestrian sample is crossed, the information such as gray scale, edge, texture, the color of target are extracted, pedestrian detection is built using learning method
Grader.Learning method mainly has adaptive enhancing iterative algorithm Adaboost, SVM and deep learning etc..
Currently, common pedestrian's feature has wavelet character etc..In addition, pedestrian is detected in the way of cascade,
Achieve preferable effect.The performance of pedestrian detection algorithm mainly by learning algorithm (grader) and Feature Descriptor the two
The restriction of factor.Feature extraction and detection process speed are all relatively slow carrying out by the methods of HOG and DPM, are not suitable for vehicle-mounted real-time
Detect environment.
Classify according to detection source, pedestrian detection can be divided into the detection based on visible images and be based on non-visible light image
Detection.Based on visible images detection often due to the complexity of pedestrian's local environment Scene, the diversity of pedestrian's appearance,
The factors such as illumination variation or climate change, influence the effect of pedestrian detection so that the accuracy rate of pedestrian detection is not high, is easy to
Cause missing inspection and flase drop.In addition, the detection based on visible light is often only applicable under daylight environment, it cannot under night condition
Enough to use, this is also one of the condition for restricting visible light and being used for intelligent driving auxiliary system.
Detection based on non-visible light image is often most commonly seen to be based on infrared image.Infrared image experience and reflection
Target and the difference of the outside radiation energy of background, describe target and background keep temperature difference condition, belong to by
Dynamic imaging, can be with all weather operations.In addition, infra-red radiation penetrates the ability of haze and air than visible light intensity.It can with gram
It takes the obstacle on partial visual and detects target object, there is stronger operating distance and anti-interference ability.Furthermore infrared figure
As having between pixel good spatial coherence, the gray average of image to keep relative stability, containing more and larger same
Matter area.Therefore, infrared image is better able to adapt to vehicle-mounted round-the-clock working environment.
But infrared imaging reflection be object heat radiation it is poor, it is very sensitive to temperature, and between object exist heat hand over
It changes, and air plays the role of scattering and absorption to heat radiation so that the contrast of infrared image is low, and object edge is fuzzy, noise
It is larger, without apparent texture and angle point.This also proposes higher challenge to the pedestrian detection based on infrared image.
In addition, pedestrian detection is further divided into the detection based on image and based on video.Compared to the pedestrian based on image
For detection, movable information and space time information in video more can provide auxiliary for pedestrian detection, to improve pedestrian
The accuracy rate of detection reduces flase drop and the missing inspection of pedestrian detection.But if the pedestrian of video is blocked, the standard of pedestrian detection
True rate is still affected.How to detect that the pedestrian being blocked in the process of moving is that this field technology urgently to be resolved hurrily is difficult
Topic.
Invention content
The purpose of the invention is to overcome the shortcomings of above-mentioned background technology, a kind of row based on vehicle mounted infrared video is provided
Detection is combined with tracking, is able to detect that the pedestrian being blocked in the process of moving, promotes row by people's detection method and system
The accuracy rate of people's detection.
The present invention provides a kind of pedestrian detection method based on vehicle mounted infrared video, includes the following steps：
S1, infrared video is extracted using integrogram and channel characteristics by vehicular infrared detection instrument acquisition infrared video
Color characteristic and Gradient Features；
S2, enhance iterative algorithm Adaboost using adaptive, the feature of step S1 extractions is trained and is screened, is examined
The pedestrian in the image of infrared video is surveyed, Preliminary detection result is obtained；
S3, using Hungary Algorithm, the Preliminary detection result that step S2 is obtained is calculated, is obtained in all images
Pedestrian is associated with the initial track sequence to be formed；
S4, the initial track sequence for obtaining step S3 calculate optimal rail as initial value using Man Kelaisi algorithms
Mark related information obtains final pursuit path.
Based on the above technical solution, step S1 specifically includes following steps：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；
On different scales, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V
(I), angle is Θ=i 6 gradient direction channel characteristics G (I) 1 [Θ=i], 1 gradient magnitude channel characteristics | | G (I) |
| it extracts, then is normalized respectively, obtain indicating the color characteristic of the frame image and the comprehensive characteristics of Gradient Features
The calculation formula of vectorial F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
Based on the above technical solution, step S2 specifically includes following steps：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input
This, yi∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, l, m, n
It is positive integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics
Weak Classifier of the training one for whether there is pedestrian in detection image, calculates the weighting fault rate of Weak Classifier, according to mistake
The Weak Classifier of rate minimum adjusts weight, and the Weak Classifier of selection is synthesized strong classifier by the selection course of iteration Weak Classifier,
For whether having pedestrian in detection image；The multi-feature vector F of all images is trained and is sieved using the strong classifier
Choosing, will detect all images of pedestrian as Preliminary detection result.
Based on the above technical solution, step S3 specifically includes following steps：
All pedestrians in the image for the Preliminary detection result that step S2 is obtained are marked with detection window, setting detection window
Similarity critical value, reject similarity be less than the critical value detection window；Using the double ratio for detecting window as geometry clue, calculate
Window is detected in the region of present frame and its correlation between the region of former frame, obtains appearance clue, in conjunction with detection window
Geometry clue and appearance clue are calculated, and the incidence matrix A between image is obtainedi,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djAfter expression
J-th of detection window of one frame image, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates one
The detection window of a target, xcorr (di,dj) indicating two maximized normalized cross correlations for detecting window, τ is to be used for
Judge the threshold value of the similarity of detection window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish all detection windows
The track of middle pedestrian obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
Based on the above technical solution, step S4 specifically includes following steps：
The initial track sequence that step S3 is obtained carries out the second secondary association as initial value, using Man Kelaisi algorithms,
Optimal data correlation is found, detection window is replaced with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIt indicates just
J-th of track in beginning track sets, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two rails
The interval of frame between mark, dist (ti,tj) indicate that the average normalization prediction of the position, width, height of corresponding track misses
Difference, xcorr ' (ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiPass between track and track
Join matrix Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained by linear regression
To final pursuit path.
The present invention also provides a kind of pedestrian detecting systems based on vehicle mounted infrared video, including feature extraction unit, training
Screening unit, the first associative cell, the second associative cell, wherein：
The feature extraction unit is used for：Infrared video is obtained by vehicular infrared detection instrument, using integrogram and channel
Feature extracts the color characteristic and Gradient Features of infrared video；
The trained screening unit is used for：Using adaptive enhancing iterative algorithm Adaboost, feature extraction unit is carried
The feature taken is trained and screens, and detects the pedestrian in the image of infrared video, obtains Preliminary detection result；
First associative cell is used for：Using Hungary Algorithm, the Preliminary detection result that training screening unit is obtained
It is calculated, obtains the pedestrian in all images and be associated with the initial track sequence to be formed；
Second associative cell is used for：The initial track sequence that first associative cell is obtained is used as initial value
Man Kelaisi algorithms calculate optimal Track association information, obtain final pursuit path.
Based on the above technical solution, the feature extraction unit is specifically used for：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；
On different scales, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V
(I), angle is Θ=i 6 gradient direction channel characteristics G (I) 1 [Θ=i], 1 gradient magnitude channel characteristics | | G (I) |
| it extracts, then is normalized respectively, obtain indicating the color characteristic of the frame image and the comprehensive characteristics of Gradient Features
The calculation formula of vectorial F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
Based on the above technical solution, the trained screening unit is specifically used for：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input
This, yi∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, l, m, n
It is positive integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics
Weak Classifier of the training one for whether there is pedestrian in detection image, calculates the weighting fault rate of Weak Classifier, according to mistake
The Weak Classifier of rate minimum adjusts weight, and the Weak Classifier of selection is synthesized strong classifier by the selection course of iteration Weak Classifier,
For whether having pedestrian in detection image；The multi-feature vector F of all images is trained and is sieved using the strong classifier
Choosing, will detect all images of pedestrian as Preliminary detection result.
Based on the above technical solution, first associative cell is specifically used for：
All pedestrians in the image for the Preliminary detection result that training screening unit obtains are marked with detection window, are arranged
The similarity critical value of window is detected, the detection window that similarity is less than the critical value is rejected；The double ratio of window will be detected as geometrical line
Rope calculates detection window in the region of present frame and its correlation between the region of former frame, appearance clue is obtained, in conjunction with inspection
The geometry clue and appearance clue for surveying window are calculated, and the incidence matrix A between image is obtainedi,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djAfter expression
J-th of detection window of one frame image, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates one
The detection window of a target, xcorr (di,dj) indicating two maximized normalized cross correlations for detecting window, τ is to be used for
Judge the threshold value of the similarity of detection window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish all detection windows
The track of middle pedestrian obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
Based on the above technical solution, second associative cell is specifically used for：
The initial track sequence that first associative cell is obtained carries out the second secondary association, using Man Kelai as initial value
This algorithm finds optimal data correlation, replaces detection window with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIt indicates just
J-th of track in beginning track sets, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two rails
The interval of frame between mark, dist (ti,tj) indicate that the average normalization prediction of the position, width, height of corresponding track misses
Difference, xcorr ' (ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiPass between track and track
Join matrix Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained by linear regression
To final pursuit path.
Compared with prior art, advantages of the present invention is as follows：
(1) present invention obtains infrared video by vehicular infrared detection instrument, and using integrogram and channel characteristics, extraction is infrared
The color characteristic and Gradient Features of video；Using adaptive enhancing iterative algorithm Adaboost, the feature of extraction is trained
And screening, the pedestrian in the image of infrared video is detected, Preliminary detection result is obtained；Using Hungary Algorithm, to Preliminary detection
As a result it is calculated, obtains the pedestrian in all images and be associated with the initial track sequence to be formed；Using initial track sequence as just
Initial value is calculated optimal Track association information, is obtained final pursuit path using Man Kelaisi algorithms.The present invention passes through row
People's detection method is detected the pedestrian target of infrared video, in combination with the side for being associated with tracking of Track association and successive frame
Method comprehensively utilizes the space time information in video, can detect the pedestrian being blocked in the process of moving, improve the row of infrared video
People's Detection accuracy.
(2) present invention uses infrared imagery technique, infrared imagery technique is a kind of imaging and passive imaging technology, is had certain
The features such as ability that anti-Changes in weather influences, and there is cloud and mist penetration capacity, all weather operations.Infrared image reflection is mesh
Be marked with and the difference of the outside radiation energy of background, describe the difference condition that target and background keeps temperature, belong to passively at
Picture, can be with all weather operations.In addition, infra-red radiation penetrates the ability of haze and air than visible light intensity.It can overcome portion
Divide visually-handicap and detect target object, there is stronger operating distance and anti-interference ability.Furthermore infrared image picture
There is good spatial coherence, the gray average of image to keep relative stability between element, contain more and larger homogeneity area.
Therefore, infrared image is better able to adapt to vehicle-mounted round-the-clock working environment, and pedestrian detection, Neng Gougeng are carried out based on infrared video
The accuracy rate of good raising pedestrian detection, increases the safety of application system.
(3) the present invention is based on videos is detected pedestrian, and the movable information and space time information in video can be more
Auxiliary information is provided for pedestrian detection, to improve the accuracy rate of pedestrian detection, reduces flase drop and the missing inspection of pedestrian detection.
(4) detection is combined with tracking by the present invention, on the basis of existing pedestrian detection result, addition track algorithm into
The optimizing detection result of one step.The method of pedestrian tracking can extract the movable information and spatial information of pedestrian in video, pass through
Detection is combined with pedestrian tracking, can effectively be applied to the space time information in video in pedestrian detection, can be effective
Improve the accuracy rate of pedestrian detection；Effectively undetected target can be carried out by successive frame association and Track association pre-
It surveys, reduces because missing inspection caused by the problems such as blocking, can be good at detecting the pedestrian being blocked in the process of moving.
(5) detection is combined by the present invention with tracking, and pedestrian detection can be more when adding track algorithm to some Weak targets
Good detection makes the detection of Weak target in infrared image imitate when carrying out road ahead pedestrian detection using vehicle mounted infrared video
Fruit is obviously improved, and the omission factor of Weak target can be effectively reduced, and further promotes the accuracy rate of pedestrian detection.
Description of the drawings
Fig. 1 is the flow chart of the pedestrian detection method based on vehicle mounted infrared video in the embodiment of the present invention.
Specific implementation mode
Below in conjunction with the accompanying drawings and specific embodiment the present invention is described in further detail.
Embodiment 1
Shown in Figure 1, the embodiment of the present invention 1 provides a kind of pedestrian detection method based on vehicle mounted infrared video, packet
Include following steps：
S1, infrared video is extracted using integrogram and channel characteristics by vehicular infrared detection instrument acquisition infrared video
Color characteristic and Gradient Features；
S2, enhance iterative algorithm Adaboost using adaptive, the feature of step S1 extractions is trained and is screened, is examined
The pedestrian in the image of infrared video is surveyed, Preliminary detection result is obtained；
S3, using Hungary Algorithm, the Preliminary detection result that step S2 is obtained is calculated, is obtained in all images
Pedestrian is associated with the initial track sequence to be formed；
S4, the initial track sequence for obtaining step S3 calculate optimal rail as initial value using Man Kelaisi algorithms
Mark related information obtains final pursuit path.
Embodiment 2
On the basis of embodiment 1, step S1 specifically includes following steps：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；
On different scales, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V
(I), angle is Θ=i 6 gradient direction channel characteristics G (I) 1 [Θ=i], 1 gradient magnitude channel characteristics | | G (I) |
| it extracts, then is normalized respectively, obtain indicating the color characteristic of the frame image and the comprehensive characteristics of Gradient Features
The calculation formula of vectorial F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
Embodiment 3
On the basis of embodiment 2, step S2 specifically includes following steps：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input
This, yi∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, l, m, n
It is positive integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics
Weak Classifier of the training one for whether there is pedestrian in detection image, calculates the weighting fault rate of Weak Classifier, according to mistake
The Weak Classifier of rate minimum adjusts weight, and the Weak Classifier of selection is synthesized strong classifier by the selection course of iteration Weak Classifier,
For whether having pedestrian in detection image；The multi-feature vector F of all images is trained and is sieved using the strong classifier
Choosing, will detect all images of pedestrian as Preliminary detection result.
Embodiment 4
On the basis of embodiment 3, step S3 specifically includes following steps：
All pedestrians in the image for the Preliminary detection result that step S2 is obtained are marked with detection window, setting detection window
Similarity critical value, reject similarity be less than the critical value detection window；Using the double ratio for detecting window as geometry clue, calculate
Window is detected in the region of present frame and its correlation between the region of former frame, obtains appearance clue, in conjunction with detection window
Geometry clue and appearance clue are calculated, and the incidence matrix A between image is obtainedi,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djAfter expression
J-th of detection window of one frame image, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates one
The detection window of a target, xcorr (di,dj) indicating two maximized normalized cross correlations for detecting window, τ is to be used for
Judge the threshold value of the similarity of detection window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish all detection windows
The track of middle pedestrian obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
Embodiment 5
On the basis of embodiment 4, step S4 specifically includes following steps：
The initial track sequence that step S3 is obtained carries out the second secondary association as initial value, using Man Kelaisi algorithms,
Optimal data correlation is found, detection window is replaced with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIt indicates just
J-th of track in beginning track sets, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two rails
The interval of frame between mark, dist (ti,tj) indicate that the average normalization prediction of the position, width, height of corresponding track misses
Difference, xcorr ' (ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiPass between track and track
Join matrix Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained by linear regression
To final pursuit path.
Embodiment 6
The embodiment of the present invention 6 provides a kind of pedestrian detecting system based on vehicle mounted infrared video, including feature extraction list
Member, training screening unit, the first associative cell, the second associative cell, wherein：
Feature extraction unit is used for：Infrared video is obtained by vehicular infrared detection instrument, using integrogram and channel characteristics,
Extract the color characteristic and Gradient Features of infrared video；
Training screening unit is used for：Using adaptive enhancing iterative algorithm Adaboost, to feature extraction unit extraction
Feature is trained and screens, and detects the pedestrian in the image of infrared video, obtains Preliminary detection result；
First associative cell is used for：Using Hungary Algorithm, the Preliminary detection result obtained to training screening unit carries out
It calculates, obtains the pedestrian in all images and be associated with the initial track sequence to be formed；
Second associative cell is used for：The initial track sequence that first associative cell is obtained is as initial value, using graceful gram
This algorithm of Lay calculates optimal Track association information, obtains final pursuit path.
Embodiment 7
On the basis of embodiment 6, feature extraction unit is specifically used for：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；
On different scales, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V
(I), angle is Θ=i 6 gradient direction channel characteristics G (I) 1 [Θ=i], 1 gradient magnitude channel characteristics | | G (I) |
| it extracts, then is normalized respectively, obtain indicating the color characteristic of the frame image and the comprehensive characteristics of Gradient Features
The calculation formula of vectorial F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
Embodiment 8
On the basis of embodiment 7, training screening unit is specifically used for：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input
This, yi∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, l, m, n
It is positive integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics
Weak Classifier of the training one for whether there is pedestrian in detection image, calculates the weighting fault rate of Weak Classifier, according to mistake
The Weak Classifier of rate minimum adjusts weight, and the Weak Classifier of selection is synthesized strong classifier by the selection course of iteration Weak Classifier,
For whether having pedestrian in detection image；The multi-feature vector F of all images is trained and is sieved using the strong classifier
Choosing, will detect all images of pedestrian as Preliminary detection result.
Embodiment 9
On the basis of embodiment 8, the first associative cell is specifically used for：
All pedestrians in the image for the Preliminary detection result that training screening unit obtains are marked with detection window, are arranged
The similarity critical value of window is detected, the detection window that similarity is less than the critical value is rejected；The double ratio of window will be detected as geometrical line
Rope calculates detection window in the region of present frame and its correlation between the region of former frame, appearance clue is obtained, in conjunction with inspection
The geometry clue and appearance clue for surveying window are calculated, and the incidence matrix A between image is obtainedi,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djAfter expression
J-th of detection window of one frame image, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates one
The detection window of a target, xcorr (di,dj) indicating two maximized normalized cross correlations for detecting window, τ is to be used for
Judge the threshold value of the similarity of detection window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish all detection windows
The track of middle pedestrian obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
Embodiment 10
On the basis of embodiment 9, the second associative cell is specifically used for：
The initial track sequence that first associative cell is obtained carries out the second secondary association, using Man Kelai as initial value
This algorithm finds optimal data correlation, replaces detection window with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIt indicates just
J-th of track in beginning track sets, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two rails
The interval of frame between mark, dist (ti,tj) indicate that the average normalization prediction of the position, width, height of corresponding track misses
Difference, xcorr ' (ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiPass between track and track
Join matrix Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained by linear regression
To final pursuit path.
The principle of the embodiment of the present invention is elaborated as follows：
The embodiment of the present invention utilizes vehicle-mounted infrared detecting set, obtains infrared video, then passes through pedestrian detection method pair
The pedestrian target of infrared video is detected.Use the row of integrating channel feature ICF (integral channel features)
People's detection method can be extracted quickly using integrogram and calculate the low-level image features such as Gradient Features and color characteristic.
Using the learning method of adaptive enhancing iterative algorithm Adaboost, the selection and study of feature are quickly carried out.
Adaboost is a kind of iterative algorithm, and core concept is to train different grader (weak typings for the same training set
Device), then these weak classifier sets are got up, constitute a stronger final classification device (strong classifier).Its algorithm itself
It is realized by change data distribution, whether it is correct according to the classification of each sample among each training set, Yi Jishang
The accuracy rate of secondary general classification, to determine the weights of each sample.Give the new data set for changing weights to sub-classification
Device is trained, and is finally finally merged the grader that each training obtains, as last Decision Classfication device.It uses
Adaboost graders can exclude some unnecessary training data features, and be placed on above crucial training data.
The specific application scenarios of pedestrian detection based on Vehicular video add on the basis of existing pedestrian detection result
Enter the further optimizing detection result of track algorithm.All companies are obtained first with Hungary Algorithm (Hungarian method)
The related information of the object detection results of continuous frame；Using obtained track as initial value, Man Kelaisi (Kuhn- are utilized
Munkres) algorithm calculates optimal Track association information, obtains final pursuit path.By the Track association of target come into
The thought of row target following can be good at detecting the pedestrian being blocked in the process of moving.
The infrared video of input is subjected to the extraction image I per frame first and calculates integrogram, using integral diagram technology to figure
Each feature channel of picture, such as Color Channel, gradient direction channel and gradient magnitude channel etc. are quickly calculated.To input picture
The scaling for carrying out image pyramid, prevents the variation of scale to the influence of feature, the extraction of each channel characteristics is different
It is carried out on scale.The feature of extraction is handled using method for normalizing.Use 3 L (I), U (I), V (I) Color Channel, G
(I) 1 [Θ=i] indicate angle be Θ=i 6 gradient direction channels and one | | G (I) | | color gradient amplitude channel,
Finally obtained feature vector can be calculate by the following formula：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
The feature of extraction is screened and trained by adaptively enhancing iterative algorithm Adaboost.If n of input
Training sample is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input, yi∈ { 0,1 } is indicated respectively
Positive sample and negative sample, wherein positive sample number are l, negative sample number m, n=l+m.Weight is initialized to each sample first, then
One Weak Classifier is trained to each feature, is characterized as that each channel for the integrating channel feature being mentioned above is special here
Sign.Then the weighting fault rate for calculating Weak Classifier adjusts weight according to best Weak Classifier (grader of error rate minimum).
The Weak Classifier of selection is synthesized strong classifier by iteration Weak Classifier selection course.
For the detection window of all picture frames of one section of sequence, it is relatively low that some scores are weeded out by a threshold value first
Detection window, remaining detection window is then done one with known track (tracklets) with Hungary Algorithm and is associated with.If certain
As soon as a detection window does not have and any path matching, then establishing a new track.
Incidence matrix is calculated in conjunction with the geometry clue and appearance clue for detecting window：Using the double ratio of detection window as geometry
Clue, appearance clue detect window by calculating and are obtained in the region of present frame and its correlation between the region of former frame,
Simultaneously in view of the uncertainty of testing result itself, a smaller threshold value (20%) can be taken, the pass between image is obtained
Join matrix Ai,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djAfter expression
J-th of detection window of one frame image, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates one
The detection window of a target, xcorr (di,dj) indicating two maximized normalized cross correlations for detecting window, τ is to be used for
Judge the threshold value of the similarity of detection window；
On the basis of obtaining related information based on Hungary Algorithm, it is used as by obtained trace information and is originally inputted progress
Secondary association：
Using Man Kelaisi algorithms, optimal data correlation is found, replaces detection window using track, while considering all
Sequence obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIt indicates just
J-th of track in beginning track sets, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two rails
The interval of frame between mark, dist (ti,tj) indicate that the average normalization prediction of the position, width, height of corresponding track misses
Difference, xcorr ' (ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiPass between track and track
Join matrix Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained by linear regression
To final pursuit path.
By the incidence matrix A between tracki,jThe high track of ' middle similarity is connected using linear regression, will be obtained final
Pursuit path.All it is by first initial track by the incidence matrix A between track as the connection that window is detected in fronti,j′
Judgement connects second initial track, becomes new track, is further continued for by the incidence matrix A between tracki,j' judge either with or without
Three initial tracks can connect, and slowly become final pursuit path in this way.It is demonstrated experimentally that this process can be good at
The influence that partial occlusion and the variable cosmetic of appropriateness etc. are brought is eliminated, longer more stable pursuit path is obtained.
It is tested by the data of acquisition, the experimental results showed that the method for the embodiment of the present invention can well solve reality
Border problem not only meets the real-time of video, can accurately also detect pedestrian.
Those skilled in the art can be carry out various modifications to the embodiment of the present invention and modification, if these modifications and change
For type within the scope of the claims in the present invention and its equivalent technologies, then these modifications and variations are also in protection scope of the present invention
Within.
The prior art that the content not being described in detail in specification is known to the skilled person.
Claims (10)
1. a kind of pedestrian detection method based on vehicle mounted infrared video, which is characterized in that include the following steps：
S1, the color of infrared video is extracted using integrogram and channel characteristics by vehicular infrared detection instrument acquisition infrared video
Feature and Gradient Features；
S2, enhance iterative algorithm Adaboost using adaptive, the feature of step S1 extractions is trained and is screened, detection is red
Pedestrian in the image of outer video obtains Preliminary detection result；
S3, using Hungary Algorithm, the Preliminary detection result that step S2 is obtained is calculated, the pedestrian in all images is obtained
It is associated with the initial track sequence formed；
S4, the initial track sequence for obtaining step S3 are calculated optimal track and are closed as initial value using Man Kelaisi algorithms
Join information, obtains final pursuit path.
2. the pedestrian detection method as described in claim 1 based on vehicle mounted infrared video, it is characterised in that：Step S1 is specifically wrapped
Include following steps：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；Not
On same scale, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V (I), folder
Angle is 6 gradient direction channel characteristics G (I) 1 [Θ=i] of Θ=i, 1 gradient magnitude channel characteristics | | G (I) | | respectively
It extracts, then is normalized, obtain indicating the color characteristic of the frame image and the multi-feature vector of Gradient Features
The calculation formula of F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
3. the pedestrian detection method as claimed in claim 2 based on vehicle mounted infrared video, it is characterised in that：Step S2 is specifically wrapped
Include following steps：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input, yi
∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, and l, m, n are just
Integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics training one
A Weak Classifier for whether there is pedestrian in detection image calculates the weighting fault rate of Weak Classifier, according to error rate minimum
Weak Classifier adjust weight, the Weak Classifier of selection is synthesized strong classifier, for examining by the selection course of iteration Weak Classifier
Whether there is pedestrian in altimetric image；The multi-feature vector F of all images is trained and is screened using the strong classifier, will be examined
All images of pedestrian are measured as Preliminary detection result.
4. the pedestrian detection method as claimed in claim 3 based on vehicle mounted infrared video, it is characterised in that：Step S3 is specifically wrapped
Include following steps：
All pedestrians in the image for the Preliminary detection result that step S2 is obtained are marked with detection window, the phase of setting detection window
Like degree critical value, the detection window that similarity is less than the critical value is rejected；Using the double ratio for detecting window as geometry clue, detection is calculated
Window obtains appearance clue in the region of present frame and its correlation between the region of former frame, in conjunction with the geometry of detection window
Clue and appearance clue are calculated, and the incidence matrix A between image is obtainedi,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djIndicate a later frame figure
J-th of detection window of picture, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates a target
Detection window, xcorr (di,dj) indicate that two maximized normalized cross correlations for detecting window, τ are for judging inspection
Survey the threshold value of the similarity of window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish row in all detection windows
The track of people obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
5. the pedestrian detection method as claimed in claim 4 based on vehicle mounted infrared video, it is characterised in that：Step S4 is specifically wrapped
Include following steps：
The initial track sequence that step S3 is obtained carries out the second secondary association as initial value, using Man Kelaisi algorithms, finds
Optimal data correlation replaces detection window with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIndicate track primary
J-th of track in mark sequence, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two tracks it
Between frame interval, dist (ti,tj) indicate that error is predicted in the average normalization of the position, width, height of corresponding track,
xcorr′(ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiAssociation square between track and track
Battle array Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained most by linear regression
Whole pursuit path.
6. a kind of pedestrian detecting system based on vehicle mounted infrared video, which is characterized in that including feature extraction unit, training screening
Unit, the first associative cell, the second associative cell, wherein：
The feature extraction unit is used for：Infrared video is obtained by vehicular infrared detection instrument, using integrogram and channel characteristics,
Extract the color characteristic and Gradient Features of infrared video；
The trained screening unit is used for：Using adaptive enhancing iterative algorithm Adaboost, to feature extraction unit extraction
Feature is trained and screens, and detects the pedestrian in the image of infrared video, obtains Preliminary detection result；
First associative cell is used for：Using Hungary Algorithm, the Preliminary detection result obtained to training screening unit carries out
It calculates, obtains the pedestrian in all images and be associated with the initial track sequence to be formed；
Second associative cell is used for：The initial track sequence that first associative cell is obtained is as initial value, using graceful gram
This algorithm of Lay calculates optimal Track association information, obtains final pursuit path.
7. the pedestrian detecting system as claimed in claim 6 based on vehicle mounted infrared video, it is characterised in that：The feature extraction
Unit is specifically used for：
Infrared video is obtained by vehicular infrared detection instrument, the scaling of image pyramid is carried out to the image of infrared video；Not
On same scale, using integrogram, to 3 Color Channel feature L (I) of each frame image I after scaling, U (I), V (I), folder
Angle is 6 gradient direction channel characteristics G (I) 1 [Θ=i] of Θ=i, 1 gradient magnitude channel characteristics | | G (I) | | respectively
It extracts, then is normalized, obtain indicating the color characteristic of the frame image and the multi-feature vector of Gradient Features
The calculation formula of F, multi-feature vector F is：
F=L (I), U (I), V (I), | | G (I) | |, G (I) 1 [Θ=i] },
8. the pedestrian detecting system as claimed in claim 7 based on vehicle mounted infrared video, it is characterised in that：The training screening
Unit is specifically used for：
If n training sample of input is：{(x1,y1),(x2,y2),…,(xn,yn), wherein xiIt is the training sample of input, yi
∈ { 0,1 } indicates positive sample and negative sample respectively, wherein positive sample number is l, and negative sample number is m, n=l+m, and l, m, n are just
Integer；Using adaptive enhancing iterative algorithm Adaboost, the weight of each sample is initialized, to each channel characteristics training one
A Weak Classifier for whether there is pedestrian in detection image calculates the weighting fault rate of Weak Classifier, according to error rate minimum
Weak Classifier adjust weight, the Weak Classifier of selection is synthesized strong classifier, for examining by the selection course of iteration Weak Classifier
Whether there is pedestrian in altimetric image；The multi-feature vector F of all images is trained and is screened using the strong classifier, will be examined
All images of pedestrian are measured as Preliminary detection result.
9. the pedestrian detecting system as claimed in claim 8 based on vehicle mounted infrared video, it is characterised in that：First association
Unit is specifically used for：
All pedestrians in the image for the Preliminary detection result that training screening unit obtains are marked with detection window, setting detection
The similarity critical value of window rejects the detection window that similarity is less than the critical value；The double ratio of window will be detected as geometry clue, meter
Detection window is calculated in the region of present frame and its correlation between the region of former frame, obtains appearance clue, in conjunction with detection window
Geometry clue and appearance clue calculated, obtain the incidence matrix A between imagei,j：
Wherein, Γ (di,dj) indicate to detect the similarity of window, diIndicate i-th of detection window of previous frame image, djIndicate a later frame figure
J-th of detection window of picture, i, j are positive integer；Γ(di,dj) numerical value is lower, similarity is higher；Box () indicates a target
Detection window, xcorr (di,dj) indicate that two maximized normalized cross correlations for detecting window, τ are for judging inspection
Survey the threshold value of the similarity of window；
Since first frame image, using Hungary Algorithm, pass through the incidence matrix A between imagei,jEstablish row in all detection windows
The track of people obtains the pedestrian in all images and is associated with the initial track sequence to be formed.
10. the pedestrian detecting system as claimed in claim 9 based on vehicle mounted infrared video, it is characterised in that：Described second closes
Receipts or other documents in duplicate member is specifically used for：
The initial track sequence that first associative cell is obtained is carried out the second secondary association, is calculated using Man Kelaisi as initial value
Method finds optimal data correlation, replaces detection window with track, obtains the incidence matrix A between tracki,j', it indicates as follows：
Γ′(ti,tj)=min (dist (ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj))；
Wherein, Γ ' (ti,tj) indicate track similarity, tiIndicate i-th of track in initial track sequence, tjIndicate track primary
J-th of track in mark sequence, Γ ' (ti,tj) numerical value is lower, the similarity of track is higher；fΔ(ti,tj) indicate two tracks it
Between frame interval, dist (ti,tj) indicate that error is predicted in the average normalization of the position, width, height of corresponding track,
xcorr′(ti,tj) indicating maximum cross correlation between track, τ ' is the threshold value for judging similarity between track；
Pass through the incidence matrix A between tracki,j' optimal Track association information is obtained, in conjunction with tiAssociation square between track and track
Battle array Ai,j' judge tjThe position of track, and and tiTrack is compared, and is obtained normalized prediction error, is obtained most by linear regression
Whole pursuit path.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810073180.4A CN108319906B (en) | 2018-01-25 | 2018-01-25 | Pedestrian detection method and system based on vehicle-mounted infrared video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810073180.4A CN108319906B (en) | 2018-01-25 | 2018-01-25 | Pedestrian detection method and system based on vehicle-mounted infrared video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108319906A true CN108319906A (en) | 2018-07-24 |
| CN108319906B CN108319906B (en) | 2020-09-29 |
Family
ID=62887825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810073180.4A Active CN108319906B (en) | 2018-01-25 | 2018-01-25 | Pedestrian detection method and system based on vehicle-mounted infrared video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108319906B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468378A (en) * | 2020-03-30 | 2021-10-01 | 阿里巴巴集团控股有限公司 | Video processing method, device and equipment and video target tracking method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110085702A1 (en) * | 2009-10-08 | 2011-04-14 | University Of Southern California | Object tracking by hierarchical association of detection responses |
| CN106203274A (en) * | 2016-06-29 | 2016-12-07 | 长沙慧联智能科技有限公司 | Pedestrian's real-time detecting system and method in a kind of video monitoring |
| CN106842183A (en) * | 2017-03-31 | 2017-06-13 | 北京无线电测量研究所 | The method and system that a kind of ripple interdigit point mark based on global nearest neighbor algorithm merges |
| CN107292911A (en) * | 2017-05-23 | 2017-10-24 | 南京邮电大学 | A kind of multi-object tracking method merged based on multi-model with data correlation |
-
2018
- 2018-01-25 CN CN201810073180.4A patent/CN108319906B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110085702A1 (en) * | 2009-10-08 | 2011-04-14 | University Of Southern California | Object tracking by hierarchical association of detection responses |
| CN106203274A (en) * | 2016-06-29 | 2016-12-07 | 长沙慧联智能科技有限公司 | Pedestrian's real-time detecting system and method in a kind of video monitoring |
| CN106842183A (en) * | 2017-03-31 | 2017-06-13 | 北京无线电测量研究所 | The method and system that a kind of ripple interdigit point mark based on global nearest neighbor algorithm merges |
| CN107292911A (en) * | 2017-05-23 | 2017-10-24 | 南京邮电大学 | A kind of multi-object tracking method merged based on multi-model with data correlation |
Non-Patent Citations (3)
| Title |
|---|
| YEONG JUN KOH 等: "CDTS: Collaborative Detection, Tracking, and Segmentation for Online Multiple Object Segmentation in Videos", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| 孙斌: "一种基于Adaboost的实时行人检测算法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 苏晓倩 等: "车载红外图像的行人检测与跟踪技术", 《激光与红外》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468378A (en) * | 2020-03-30 | 2021-10-01 | 阿里巴巴集团控股有限公司 | Video processing method, device and equipment and video target tracking method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108319906B (en) | 2020-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| Kwak et al. | Pedestrian tracking using online boosted random ferns learning in far-infrared imagery for safe driving at night | |
| Huang et al. | On-board vision system for lane recognition and front-vehicle detection to enhance driver's awareness | |
| CN103198332B (en) | A kind of far infrared vehicle-mounted pedestrian detection method of real-time robust | |
| CN104183127B (en) | Traffic surveillance video detection method and device | |
| US20120045119A1 (en) | Method of identifying an object in a visual scene | |
| CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
| CN102609720B (en) | Pedestrian detection method based on position correction model | |
| CN105512720A (en) | Public transport vehicle passenger flow statistical method and system | |
| CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
| CN103902976A (en) | Pedestrian detection method based on infrared image | |
| CN103390164A (en) | Object detection method based on depth image and implementing device thereof | |
| CN105404857A (en) | Infrared-based night intelligent vehicle front pedestrian detection method | |
| CN103049751A (en) | Improved weighting region matching high-altitude video pedestrian recognizing method | |
| CN102855500A (en) | Haar and HoG characteristic based preceding car detection method | |
| CN103617426A (en) | Pedestrian target detection method under interference by natural environment and shelter | |
| CN105930803A (en) | Preceding vehicle detection method based on Edge Boxes and preceding vehicle detection device thereof | |
| CN109948690A (en) | A kind of high-speed rail scene perception method based on deep learning and structural information | |
| CN106570490A (en) | Pedestrian real-time tracking method based on fast clustering | |
| CN110188606A (en) | Lane recognition method, device and electronic equipment based on high light spectrum image-forming | |
| CN103034843A (en) | Method for detecting vehicle at night based on monocular vision | |
| CN104517095A (en) | Head division method based on depth image | |
| Elihos et al. | Comparison of image classification and object detection for passenger seat belt violation detection using NIR & RGB surveillance camera images | |
| CN105913026A (en) | Passenger detecting method based on Haar-PCA characteristic and probability neural network | |
| CN107085703A (en) | Merge face detection and the automobile passenger method of counting of tracking |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240205 Address after: No. 548, 5th Floor, Building 10, No. 28 Linping Avenue, Donghu Street, Linping District, Hangzhou City, Zhejiang Province Patentee after: Hangzhou Tuke Intelligent Information Technology Co.,Ltd. Country or region after: China Address before: 430074 No. 182, National Road, Hongshan District, Wuhan, Hubei Patentee before: SOUTH CENTRAL University FOR NATIONALITIES Country or region before: China |
|
| TR01 | Transfer of patent right |