CN108319906B - Pedestrian detection method and system based on vehicle-mounted infrared video - Google Patents
Pedestrian detection method and system based on vehicle-mounted infrared video Download PDFInfo
- Publication number
- CN108319906B CN108319906B CN201810073180.4A CN201810073180A CN108319906B CN 108319906 B CN108319906 B CN 108319906B CN 201810073180 A CN201810073180 A CN 201810073180A CN 108319906 B CN108319906 B CN 108319906B
- Authority
- CN
- China
- Prior art keywords
- detection
- tracks
- track
- infrared video
- representing
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 212
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000000034 method Methods 0.000 claims abstract description 32
- 238000012216 screening Methods 0.000 claims abstract description 24
- 230000008569 process Effects 0.000 claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims description 38
- 238000000605 extraction Methods 0.000 claims description 14
- 230000003044 adaptive effect Effects 0.000 claims description 9
- 238000012417 linear regression Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 241001672694 Citrus reticulata Species 0.000 claims 1
- 230000008859 change Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000005855 radiation Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000003331 infrared imaging Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000003595 mist Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
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
Abstract
The invention discloses a pedestrian detection method and system based on an on-vehicle infrared video, and relates to the field of computer vision of pedestrian detection. The method comprises the following steps: acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics; training and screening the extracted features by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the image of the infrared video to obtain a primary detection result; calculating the preliminary detection result by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images; and taking the initial track sequence as an initial value, and calculating the optimal track association information by adopting a ManCleis algorithm to obtain the final tracking track. The invention combines detection and tracking, can detect the shielded pedestrians in the driving process, and improves the accuracy of pedestrian detection.
Description
Technical Field
The invention relates to the field of computer vision for pedestrian detection, in particular to a pedestrian detection method and system based on vehicle-mounted infrared video.
Background
The pedestrian detection can be classified into a motion characteristic-based method, a template matching-based method, and a statistical learning-based method according to the method of detection. Compared with the former two pedestrian detection methods, the statistical learning-based method has the advantages of high detection precision and better robustness, and is a key point and a hot point of the current pedestrian detection research. The statistical learning-based method extracts information such as gray scale, edge, texture, color and the like of a target through a large number of pedestrian samples, and a pedestrian detection classifier is constructed by utilizing a learning method. The learning method mainly comprises an adaptive enhancement iterative algorithm Adaboost, a support vector machine, deep learning and the like.
At present, the pedestrian features commonly used include wavelet features and the like. In addition, the pedestrian is detected in a cascading mode, and a good effect is achieved. The performance of the pedestrian detection algorithm is mainly limited by two factors, a learning algorithm (classifier) and a feature descriptor. The methods such as HOG and DPM are slow in the process of feature extraction and detection, and are not suitable for vehicle-mounted real-time detection environments.
The pedestrian detection may be classified into detection based on a visible light image and detection based on a non-visible light image according to the detection source classification. Based on visible light image detection, the effect of pedestrian detection is often influenced by factors such as complexity of scenes in the environment where pedestrians are located, diversity of appearances of pedestrians, illumination change or climate change, so that the accuracy of pedestrian detection is low, and missing detection and false detection are easily caused. In addition, detection based on visible light is often only applicable in daytime environments and not applicable in night conditions, which is one of the conditions that restrict the use of visible light for intelligent driving assistance systems.
Detection based on non-visible light images is most often based on infrared images. The infrared image senses and reflects the difference of the energy radiated outwards by the target and the background, describes the difference condition of the temperature kept by the target and the background, belongs to passive imaging, and can work all the day. In addition, the ability of infrared radiation to penetrate haze and the atmosphere is stronger than visible light. The device can overcome partial visual obstacles to detect the target object, and has strong action distance and anti-interference capability. Moreover, the infrared image has good spatial correlation among pixels, the gray average value of the image is kept relatively stable, and the image contains more and larger homogeneous regions. Therefore, the infrared image can be more suitable for the vehicle-mounted all-weather working environment.
However, infrared imaging reflects the difference in thermal radiation of objects, which is very sensitive to temperature, and heat exchange exists between objects, and air has scattering and absorbing effects on thermal radiation, so that the infrared image has low contrast, blurred edges of objects, high noise, and no obvious texture or corner points. This also presents a higher challenge to pedestrian detection based on infrared images.
In addition, pedestrian detection can be divided into image-based and video-based detection. Compared with pedestrian detection based on images, the motion information and the spatiotemporal information in the video can provide more assistance for the pedestrian detection, so that the accuracy of the pedestrian detection is improved, and the false detection and the missing detection of the pedestrian detection are reduced. However, if the pedestrian of the video is occluded, the accuracy of the pedestrian detection is still affected. How to detect the blocked pedestrian in the driving process is an urgent technical problem to be solved in the field.
Disclosure of Invention
The invention aims to overcome the defects of the background technology and provide a pedestrian detection method and system based on an on-vehicle infrared video, which combines detection and tracking, can detect the blocked pedestrians in the driving process and improves the accuracy of pedestrian detection.
The invention provides a pedestrian detection method based on a vehicle-mounted infrared video, which comprises the following steps of:
s1, acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
s2, training and screening the features extracted in the step S1 by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the images of the infrared video to obtain a primary detection result;
s3, calculating the preliminary detection result obtained in the step S2 by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images;
and S4, taking the initial track sequence obtained in the step S3 as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain a final tracking track.
On the basis of the above technical solution, step S1 specifically includes the following steps:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
on the basis of the above technical solution, step S2 specifically includes the following steps:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
On the basis of the above technical solution, step S3 specifically includes the following steps:
marking all pedestrians in the image of the primary detection result obtained in the step S2 by using a detection window, setting a similarity critical value of the detection window, and rejecting the detection window with the similarity lower than the critical value; will detectThe cross ratio of the window is used as a geometric clue, the correlation between the area of the detection window in the current frame and the area of the detection window in the previous frame is calculated to obtain an appearance clue, and the geometric clue and the appearance clue of the detection window are combined for calculation to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jAnd establishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images.
On the basis of the above technical solution, step S4 specifically includes the following steps:
taking the initial track sequence obtained in the step S3 as an initial value, performing second correlation, searching for optimal data correlation by adopting a ManCleis algorithm, and replacing a detection window with a track to obtain a correlation matrix A between tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
The invention also provides a pedestrian detection system based on the vehicle-mounted infrared video, which comprises a feature extraction unit, a training screening unit, a first association unit and a second association unit, wherein:
the feature extraction unit is configured to: acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
the training screening unit is used for: training and screening the features extracted by the feature extraction unit by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the image of the infrared video to obtain a primary detection result;
the first association unit is configured to: calculating the primary detection result obtained by the training screening unit by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images;
the second association unit is configured to: and taking the initial track sequence obtained by the first association unit as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain the final tracking track.
On the basis of the above technical solution, the feature extraction unit is specifically configured to:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
on the basis of the technical scheme, the training screening unit is specifically used for:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
On the basis of the above technical solution, the first association unit is specifically configured to:
the initial value obtained by training the screening unitMarking all pedestrians in the image of the detection result by using a detection window, setting a similarity critical value of the detection window, and removing the detection window with the similarity lower than the critical value; taking the cross ratio of the detection window as a geometric clue, calculating the correlation between the region of the detection window in the current frame and the region of the detection window in the previous frame to obtain an appearance clue, and calculating by combining the geometric clue and the appearance clue of the detection window to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jAnd establishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images.
On the basis of the above technical solution, the second association unit is specifically configured to:
taking the initial track sequence obtained by the first association unit as an initial value, performing second association, searching for optimal data association by adopting a ManCleis algorithm, and replacing a detection window with a track to obtain an association matrix A between tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
Compared with the prior art, the invention has the following advantages:
(1) acquiring an infrared video by a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics; training and screening the extracted features by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the image of the infrared video to obtain a primary detection result; calculating the preliminary detection result by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images; and taking the initial track sequence as an initial value, and calculating the optimal track association information by adopting a ManCleis algorithm to obtain the final tracking track. The pedestrian detection method detects the pedestrian target of the infrared video, combines the track association and the continuous frame association tracking method, comprehensively utilizes the space-time information in the video, can detect the blocked pedestrian in the driving process, and improves the pedestrian detection accuracy of the infrared video.
(2) The invention uses an infrared imaging technology, which is a passive imaging technology, has certain capability of resisting the influence of weather change, and has the characteristics of cloud and mist penetration capability, all-weather work and the like. The infrared image reflects the difference of the energy radiated outwards by the target and the background, describes the difference condition of the temperature maintained by the target and the background, belongs to passive imaging, and can work all the day. In addition, the ability of infrared radiation to penetrate haze and the atmosphere is stronger than visible light. The device can overcome partial visual obstacles to detect the target object, and has strong action distance and anti-interference capability. Moreover, the infrared image has good spatial correlation among pixels, the gray average value of the image is kept relatively stable, and the image contains more and larger homogeneous regions. Therefore, the infrared image can be more suitable for the vehicle-mounted all-weather working environment, the pedestrian detection is carried out based on the infrared video, the accuracy of the pedestrian detection can be better improved, and the safety of an application system is improved.
(3) The pedestrian detection method based on the video detects the pedestrian, and the motion information and the space-time information in the video can provide more auxiliary information for the pedestrian detection, so that the accuracy of the pedestrian detection is improved, and the false detection and the missing detection of the pedestrian detection are reduced.
(4) The invention combines detection and tracking, and adds a tracking algorithm to further optimize the detection result on the basis of the existing pedestrian detection result. The pedestrian tracking method can extract the motion information and the spatial information of the pedestrian in the video, and by combining detection and pedestrian tracking, the spatiotemporal information in the video can be effectively applied to pedestrian detection, so that the accuracy of the pedestrian detection can be effectively improved; the undetected target can be effectively predicted through continuous frame association and track association, missing detection caused by shielding and other problems is reduced, and the shielded pedestrians can be well detected in the driving process.
(5) The method combines detection and tracking, the pedestrian detection and tracking algorithm can better detect some weak and small targets, and when the vehicle-mounted infrared video is used for detecting pedestrians on the road ahead, the detection effect of the weak and small targets in the infrared image is obviously improved, the omission ratio of the weak and small targets can be effectively reduced, and the accuracy of pedestrian detection is further improved.
Drawings
Fig. 1 is a flowchart of a pedestrian detection method based on an onboard infrared video in an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the embodiments.
Example 1
Referring to fig. 1, an embodiment 1 of the present invention provides a pedestrian detection method based on an onboard infrared video, including the following steps:
s1, acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
s2, training and screening the features extracted in the step S1 by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the images of the infrared video to obtain a primary detection result;
s3, calculating the preliminary detection result obtained in the step S2 by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images;
and S4, taking the initial track sequence obtained in the step S3 as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain a final tracking track.
Example 2
On the basis of embodiment 1, step S1 specifically includes the following steps:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
example 3
On the basis of embodiment 2, step S2 specifically includes the following steps:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
Example 4
On the basis of embodiment 3, step S3 specifically includes the following steps:
marking all pedestrians in the image of the primary detection result obtained in the step S2 by using a detection window, setting a similarity critical value of the detection window, and rejecting the detection window with the similarity lower than the critical value; taking the cross ratio of the detection window as a geometric clue, calculating the correlation between the region of the detection window in the current frame and the region of the detection window in the previous frame to obtain an appearance clue, and calculating by combining the geometric clue and the appearance clue of the detection window to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jAnd establishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images.
Example 5
On the basis of embodiment 4, step S4 specifically includes the following steps:
taking the initial track sequence obtained in the step S3 as an initial value, performing second correlation, searching for optimal data correlation by adopting a ManCleis algorithm, and replacing a detection window with a track to obtain a correlation matrix A between tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
Example 6
The embodiment 6 of the invention provides a pedestrian detection system based on a vehicle-mounted infrared video, which comprises a feature extraction unit, a training screening unit, a first association unit and a second association unit, wherein:
the feature extraction unit is configured to: acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
the training screening unit is used for: training and screening the features extracted by the feature extraction unit by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the image of the infrared video to obtain a primary detection result;
the first association unit is used for: calculating the primary detection result obtained by the training screening unit by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images;
the second association unit is used for: and taking the initial track sequence obtained by the first association unit as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain the final tracking track.
Example 7
On the basis of embodiment 6, the feature extraction unit is specifically configured to:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
example 8
On the basis of example 7, the training screening unit is specifically configured to:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
Example 9
On the basis of embodiment 8, the first association unit is specifically configured to:
marking all pedestrians in the image of the primary detection result obtained by the training and screening unit by using a detection window, setting a similarity critical value of the detection window, and rejecting the detection window with the similarity lower than the critical value; using the cross ratio of the detection window as a geometric clueCalculating the correlation between the current frame region and the previous frame region of the detection window to obtain an appearance clue, and calculating by combining the geometric clue and the appearance clue of the detection window to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jAnd establishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images.
Example 10
On the basis of embodiment 9, the second association unit is specifically configured to:
taking the initial track sequence obtained by the first association unit as an initial value, performing second association, searching for optimal data association by adopting a ManCleis algorithm, and replacing a detection window with a track to obtain an association matrix A between tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
The principle of the embodiment of the invention is elaborated as follows:
according to the embodiment of the invention, the vehicle-mounted infrared detector is used for acquiring the infrared video, and then the pedestrian target of the infrared video is detected by the pedestrian detection method. The pedestrian detection method using the integrated channel features ICF (integrated channel features) can rapidly extract and calculate the bottom layer features such as gradient features and color features by using the integral map.
And (3) rapidly selecting and learning the characteristics by using a learning method of an adaptive enhancement iterative algorithm Adaboost. Adaboost is an iterative algorithm, and the core idea thereof is to train different classifiers (weak classifiers) aiming at the same training set, and then to assemble the weak classifiers to form a stronger final classifier (strong classifier). The algorithm is realized by changing data distribution, and determines the weight of each sample according to whether the classification of each sample in each training set is correct and the accuracy of the last overall classification. And (4) sending the new data set with the modified weight value to a lower-layer classifier for training, and finally fusing the classifiers obtained by each training as a final decision classifier. The use of the adaboost classifier may exclude some unnecessary training data features and place them on top of the critical training data.
Based on a specific application scene of pedestrian detection of the vehicle-mounted video, a tracking algorithm is added to further optimize a detection result on the basis of an existing pedestrian detection result. Firstly, obtaining the associated information of target detection results of all continuous frames by using Hungary algorithm (Hungarian method); and taking the obtained track as an initial value, and calculating the optimal track association information by utilizing a Mandarin-Munkres algorithm to obtain the final tracking track. The idea of tracking the target by the track association of the target can well detect the blocked pedestrian in the driving process.
Firstly, each frame of an input infrared video is carried out to extract an image I and calculate an integral map, and each characteristic channel of the image, such as a color channel, a gradient direction channel, a gradient amplitude channel and the like, is rapidly calculated by utilizing the integral map technology. And scaling the image pyramid of the input image to prevent the influence of the change of the scale on the features, wherein the extraction of each channel feature is performed on different scales. The extracted features are processed using a normalization method. Using 3 l (i), u (i), v (i) color channels, g (i) · 1[ Θ ═ i ] denotes 6 gradient direction channels with an included angle Θ ═ i and an | g (i) | | color gradient magnitude channel, the resulting feature vector can be calculated by the following formula:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
and screening and training the extracted features by using an adaptive enhancement iterative algorithm Adaboost. Let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} denotes positive and negative samples, respectively, where the number of positive samples is l, the number of negative samples is m, n ═ l + mThe weights are initialized and a weak classifier is then trained on each feature, here each of the above-mentioned integrated channel features. The weighted error rate of the weak classifiers is then calculated and the weights are adjusted according to the best weak classifier (the classifier with the lowest error rate). And iterating the weak classifier selection process, and synthesizing the selected weak classifiers into a strong classifier.
For the detection windows of all image frames of a sequence, some detection windows with lower scores are first rejected by a threshold value, and then the remaining detection windows are associated with known trajectories (tracklets) by the hungarian algorithm. If a detection window does not match any of the tracks, a new track is established.
And calculating the incidence matrix by combining the geometric clues and the appearance clues of the detection window: the cross ratio of the detection window is used as a geometric clue, an appearance clue is obtained by calculating the correlation between the area of the detection window in the current frame and the area of the detection window in the previous frame, and a relatively small threshold value (20 percent) can be selected to obtain an association matrix A between images in consideration of the uncertainty of the detection result per sei,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
on the basis of obtaining the association information based on the Hungarian algorithm, performing second association by taking the obtained track information as an original input:
the optimal data association is searched by adopting a Manclais algorithm, the tracks are used for replacing a detection window, and all sequences are considered at the same time to obtain an association matrix A between the tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
Relating matrix A between tracksi,j' the tracks with high similarity are connected by linear regression, and the final tracking track is obtained. As with the previous detection window connections, the first initial trace is derived from the correlation matrix A between the tracesi,j' judge and connect the second initial track, change to the new track, and continue with the correlation matrix A between the tracksi,j' judging whether a third initial track can be connected or not, and slowly changing into a final tracking track. Experiments prove thatThe process can well eliminate the influence caused by local shielding, proper appearance transformation and the like, and obtain a longer and more stable tracking track.
The experiment is carried out through the collected data, and the experiment result shows that the method provided by the embodiment of the invention can well solve the practical problem, not only meets the real-time performance of the video, but also can accurately detect the pedestrian.
Various modifications and variations of the embodiments of the present invention may be made by those skilled in the art, and they are also within the scope of the present invention, provided they are within the scope of the claims of the present invention and their equivalents.
What is not described in detail in the specification is prior art that is well known to those skilled in the art.
Claims (8)
1. A pedestrian detection method based on vehicle-mounted infrared video is characterized by comprising the following steps:
s1, acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
s2, training and screening the features extracted in the step S1 by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the images of the infrared video to obtain a primary detection result;
s3, calculating the preliminary detection result obtained in the step S2 by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images, and specifically comprising the following steps: marking all pedestrians in the image of the primary detection result obtained in the step S2 by using a detection window, setting a similarity critical value of the detection window, and rejecting the detection window with the similarity lower than the critical value; taking the cross ratio of the detection window as a geometric clue, calculating the correlation between the region of the detection window in the current frame and the region of the detection window in the previous frame to obtain an appearance clue, and calculating by combining the geometric clue and the appearance clue of the detection window to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jEstablishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images;
and S4, taking the initial track sequence obtained in the step S3 as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain a final tracking track.
2. The pedestrian detection method based on the vehicle-mounted infrared video according to claim 1, characterized in that: step S1 specifically includes the following steps:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
3. the pedestrian detection method based on the vehicle-mounted infrared video according to claim 2, characterized in that: step S2 specifically includes the following steps:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
4. The pedestrian detection method based on the vehicle-mounted infrared video according to claim 1, characterized in that: step S4 specifically includes the following steps:
taking the initial track sequence obtained in the step S3 as an initial value, performing second correlation, searching for optimal data correlation by adopting a ManCleis algorithm, and replacing a detection window with a track to obtain a correlation matrix A between tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
5. The utility model provides a pedestrian detection system based on-vehicle infrared video which characterized in that, includes feature extraction unit, training screening unit, first associative unit, second associative unit, wherein:
the feature extraction unit is configured to: acquiring an infrared video through a vehicle-mounted infrared detector, and extracting color characteristics and gradient characteristics of the infrared video by adopting an integral map and channel characteristics;
the training screening unit is used for: training and screening the features extracted by the feature extraction unit by adopting a self-adaptive enhancement iterative algorithm Adaboost, and detecting pedestrians in the image of the infrared video to obtain a primary detection result;
the first association unit is configured to: the method comprises the following steps of calculating a primary detection result obtained by a training screening unit by adopting a Hungarian algorithm to obtain an initial track sequence formed by association of pedestrians in all images, and specifically comprises the following steps: marking all pedestrians in the image of the primary detection result obtained by the training and screening unit by using the detection window, setting the similarity critical value of the detection window, and rejecting similar pedestriansA detection window with a degree lower than the critical value; taking the cross ratio of the detection window as a geometric clue, calculating the correlation between the region of the detection window in the current frame and the region of the detection window in the previous frame to obtain an appearance clue, and calculating by combining the geometric clue and the appearance clue of the detection window to obtain an incidence matrix A between the imagesi,j:
Wherein (d)i,dj) Representing the degree of similarity of the detection windows, diRepresenting the ith detection window of the previous frame image, djA j detection window representing a next frame image, wherein i and j are positive integers; (d)i,dj) The lower the numerical value, the higher the similarity; box (-) denotes the detection window of an object, xcorr (d)i,dj) Represents the maximized normalized cross-correlation of the two detection windows, τ being the threshold for judging the similarity of the detection windows;
starting from the first frame image, adopting Hungarian algorithm and passing through the incidence matrix A between the imagesi,jEstablishing the tracks of the pedestrians in all the detection windows to obtain an initial track sequence formed by association of the pedestrians in all the images;
the second association unit is configured to: and taking the initial track sequence obtained by the first association unit as an initial value, and calculating the optimal track association information by adopting a Manchris algorithm to obtain the final tracking track.
6. The vehicle-mounted infrared video-based pedestrian detection system of claim 5, wherein: the feature extraction unit is specifically configured to:
acquiring an infrared video through a vehicle-mounted infrared detector, and zooming an image pyramid of an image of the infrared video; on different scales, an integral graph is adopted, 3 color channel features L (I), U (I), V (I) and 6 gradient direction channel features G (I) & 1[ theta ═ I ] with an included angle theta ═ I and 1 gradient amplitude channel feature | | | | G (I) | | | of each frame image I are extracted respectively, normalization processing is carried out, a comprehensive feature vector F representing the color features and the gradient features of the frame image is obtained, and the calculation formula of the comprehensive feature vector F is as follows:
F={L(I),U(I),V(I),||G(I)||,G(I)·1[Θ=i]},
7. the vehicle-mounted infrared video-based pedestrian detection system of claim 6, wherein: the training screening unit is specifically configured to:
let the input n training samples be: { (x)1,y1),(x2,y2),…,(xn,yn) In which xiIs an input training sample, yi∈ {0,1} respectively represents positive samples and negative samples, wherein the number of the positive samples is l, the number of the negative samples is m, n is l + m, l, m and n are positive integers, the adaptive enhancement iterative algorithm Adaboost is adopted to initialize the weight of each sample, a weak classifier for detecting whether a pedestrian exists in an image is trained on each channel characteristic, the weighted error rate of the weak classifier is calculated, the weight is adjusted according to the weak classifier with the minimum error rate, the selection process of the weak classifier is iterated, the selected weak classifiers are combined into a strong classifier for detecting whether the pedestrian exists in the image, the strong classifier is adopted to train and screen the comprehensive characteristic vector F of all the images, and all the images with the detected pedestrian are used as preliminary detection results.
8. The vehicle-mounted infrared video-based pedestrian detection system of claim 5, wherein: the second association unit is specifically configured to:
taking the initial track sequence obtained by the first association unit as an initial value, performing second association, adopting a Mandarin algorithm to search for optimal data association, and replacing tracks with tracksDetecting the window to obtain the incidence matrix A between the tracksi,j', denotes the following:
′(ti,tj)=min(dist(ti,tj),dist(ti,tj))×(1-xcorr′(ti,tj));
wherein' (t)i,tj) Representing the degree of similarity of the tracks, tiRepresenting the ith track, t, in the initial sequence of tracksjDenotes the jth track,' (t) in the initial track sequencei,tj) The lower the numerical value, the higher the similarity of the tracks; f. ofΔ(ti,tj) Indicates the interval of frames between two tracks, dist (t)i,tj) Normalized prediction error, xcorr' (t), representing the average of the position, width, height of the corresponding trajectoryi,tj) Representing the maximum cross correlation among the tracks, wherein tau' is a threshold value used for judging the similarity among the tracks;
by correlation matrix A between tracksi,j' obtaining the optimal track association information, combining with tiTracks and correlation matrix A between tracksi,j' judgment tjPosition of the track, andiand comparing the tracks to obtain a normalized prediction error, and obtaining a final tracking track through linear regression.
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 CN108319906A (en) | 2018-07-24 |
CN108319906B true 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) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8705793B2 (en) * | 2009-10-08 | 2014-04-22 | University Of Southern California | Object tracking by hierarchical association of detection responses |
-
2018
- 2018-01-25 CN CN201810073180.4A patent/CN108319906B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
---|
CDTS: Collaborative Detection, Tracking, and Segmentation for Online Multiple Object Segmentation in Videos;Yeong Jun Koh 等;《2017 IEEE International Conference on Computer Vision (ICCV)》;20171031;第3601-3609页,第3.1节,图2 * |
一种基于Adaboost的实时行人检测算法;孙斌;《中国优秀硕士学位论文全文数据库 信息科技辑》;20151215(第12期);第I138-872页,正文第24-25、32-33、36-40页,图 3-1 * |
车载红外图像的行人检测与跟踪技术;苏晓倩 等;《激光与红外》;20120831;第42卷(第8期);第949-953页,第1-4节 * |
Also Published As
Publication number | Publication date |
---|---|
CN108319906A (en) | 2018-07-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108983219B (en) | Fusion method and system for image information and radar information of traffic scene | |
Kwak et al. | Pedestrian tracking using online boosted random ferns learning in far-infrared imagery for safe driving at night | |
John et al. | Pedestrian detection in thermal images using adaptive fuzzy C-means clustering and convolutional neural networks | |
CN111914664A (en) | Vehicle multi-target detection and track tracking method based on re-identification | |
CN108549864B (en) | Vehicle-mounted thermal imaging pedestrian detection-oriented region-of-interest filtering method and device | |
Keller et al. | The benefits of dense stereo for pedestrian detection | |
US7639841B2 (en) | System and method for on-road detection of a vehicle using knowledge fusion | |
CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
Elzein et al. | A motion and shape-based pedestrian detection algorithm | |
Ogale | A survey of techniques for human detection from video | |
Liu et al. | A novel distance estimation method leading a forward collision avoidance assist system for vehicles on highways | |
Zin et al. | Fusion of infrared and visible images for robust person detection | |
Hsu et al. | Passenger flow counting in buses based on deep learning using surveillance video | |
CN108564598B (en) | Improved online Boosting target tracking method | |
Miller et al. | Person tracking in UAV video | |
Yang et al. | On-road collision warning based on multiple FOE segmentation using a dashboard camera | |
CN110837769A (en) | Embedded far infrared pedestrian detection method based on image processing and deep learning | |
Zhu et al. | Pedestrian detection in low-resolution imagery by learning multi-scale intrinsic motion structures (mims) | |
Gerónimo et al. | Pedestrian detection using adaboost learning of features and vehicle pitch estimation | |
Hou et al. | Human detection and tracking over camera networks: A review | |
CN108319906B (en) | Pedestrian detection method and system based on vehicle-mounted infrared video | |
Zhang et al. | An efficient and flexible approach for multiple vehicle tracking in the aerial video sequence | |
O'Malley et al. | A review of automotive infrared pedestrian detection techniques | |
Wu et al. | Real-time running detection from a moving camera | |
Qadar et al. | A comparative study of nighttime object detection with datasets from australia and china |
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 |