CN103617426A - Pedestrian target detection method under interference by natural environment and shelter - Google Patents
Pedestrian target detection method under interference by natural environment and shelter Download PDFInfo
- Publication number
- CN103617426A CN103617426A CN201310643883.3A CN201310643883A CN103617426A CN 103617426 A CN103617426 A CN 103617426A CN 201310643883 A CN201310643883 A CN 201310643883A CN 103617426 A CN103617426 A CN 103617426A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- neu
- target
- sample
- image
- 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
Images
Abstract
The invention discloses a pedestrian target detection method under interference by the natural environment and a shelter. The HOG-CLBP feature extraction method is provided and combined with an NEU-Person sample library and a human body part model obtained through training of the NEU-Person sample library, the HOG-CLBP feature extraction method, the NEU-Person sample library and the human body part model are matched with a SVM classifier for converting a pedestrian detection problem into a double classification problem to carry out detection. Positive and negative samples which are trained in a concentrated mode are randomly extracted from the two libraries for training and test samples are randomly extracted from the two libraries to carry out testing. The positive samples are positive samples in a test set in the NEU-Person sample library. The NEU-Person sample library is used for carrying out training to generate a part model for detection. The NEU-Person sample library relates to samples of various environments such as rain, snow, fog, night and camouflage, thus the robustness of pedestrian detection is quite strong.
Description
Technical field
The present invention relates to a kind of physical environment and disturb and have the pedestrian target detection method of blocking, is a kind of high robust pedestrian detection method based on machine vision study specifically, can be used for Complex Natural Environment and has the pedestrian detection under circumstance of occlusion.
Background technology
It is that to realize mechanized equipment visual that pedestrian target detects, intelligent and unmanned, also be public safety simultaneously, the important foundation of military surveillance, it is the study hotspot in computer vision field, but pedestrian's state exists residing environment different, a clothing great variety of models of wearing, human body figure's difference, during Information Monitoring, the different situation such as the attitude of human body and the camera shooting angle while gathering is larger on detecting the impact of effect, need the stronger detection algorithm of high robust of detection efficiency, therefore how can detect effectively and accurately pedestrian becomes challenging problem.
Now widely used pedestrian's Feature Descriptor is HOG (the Histograms of Oriented Gradients) method being proposed by people such as Dalal, and this detection method is that the window calculation gradient orientation histogram that carries out intensive scanning by treating detected image obtains.The most important thought of HOG describer is: in a sub-picture, the presentation of localized target and shape (appearance and shape) can be described well by the direction Density Distribution at gradient or edge.Therefore HOG can depict gradient information and the profile information of human body effectively.But HOG intrinsic dimensionality is higher, calculate slowly.
Local binary patterns (Local Binary Pattern, LBP) is the texture measure in a kind of tonal range, at first by people such as Ojala T. for the local contrast of complementary ground dimensioned plan picture proposes, be now widely used in recognition of face.LBP operator is derived from the local contiguous definition of texture, it is a kind of constant texture analysis method of gray level, its advantage is: the variation of intensity of illumination is had to stronger robustness, computing velocity is fast, gray feature to monotone variation has unchangeability, characterize preferably local message and texture information, but its discriminating power in low resolution situation is poor.
All there is the limitation of self in the feature that single method is extracted, for these deficiencies, the HOG-LBP method of feature is extracted HOG and the combination of LPB method in people's propositions such as Xiaoyu Wang, both comprised abundant texture information and the colouring information (LBP) of human body, also profile information and the local gradient information (HOG) of human body have well been characterized, in INRIA pedestrian's database test, obtained good experimental result, but the blocking of different proportion, visibility in various degree Complex Natural Environment and the camouflage in various degree under condition detect Shortcomings.
Summary of the invention
The object of the invention is to be the problem for above-mentioned prior art existence, provide a kind of physical environment to disturb and have the pedestrian target detection method while blocking.
To achieve these goals, the present invention adopts following technical scheme:
Physical environment disturbs and has the pedestrian target detection method while blocking, and the method comprises the steps:
Step 1, set up pedestrian's training sample set, sample set is divided into the positive sample set that has pedestrian and the negative sample collection that there is no pedestrian; Sample is from NEU-Person Sample Storehouse and INRIA pedestrian's database;
Step 2, employing HOG-CLBP feature extracting method extract the characteristics of human body in sample set, by piecemeal, process image, the feature of extracting is carried out to PCA dimensionality reduction, obtain the global feature of pedestrian's training sample set sample;
Step 3, the feature of extraction is sent into svm classifier device classify, obtain the pedestrian's sorter based on NEU-Person Sample Storehouse;
Step 4, the pedestrian who extracts NEU-Person Sample Storehouse, calculate the approximate weighted value of all parts of pedestrian in NEU-person Sample Storehouse;
Step 5, complete laggard every trade people detection of training stage, with intensive scanning, detect testing image, by sliding window, operate, utilize the sorter training scan each window and adjudicate, the window of judging pedestrian carries out mark, finally the markd window of institute is merged;
Step 6, by image pyramid and multiple dimensioned sliding window, operate, the rectangle frame under the different yardstick of image is merged into a frame, and then obtains target position accurately, be about to pedestrian detection to be measured out.
Advantage of the present invention is:
1, the present invention utilizes the method for HOG-CLBP combination to extract feature, has fully taken into account upright human body and has not only had a large amount of profile gradient informations, and also have a large amount of texture informations on outward appearance and clothing, can identify better pedestrian.
2, NEU-Person Sample Storehouse of the present invention train produce for detection of partial model, NEU-Person Sample Storehouse relates to the sample of the multiple environment such as rain, snow, mist, night and camouflage color, therefore makes pedestrian detection have very strong robustness.
3, the present invention, through PCA dimensionality reduction and multiscale analysis, by training and the study of positive negative sample, produces the pedestrian's feature for classifying, and has so just reduced training and has detected the needed time, improves accuracy of detection.
4, the present invention uses svm classifier device to carry out classification and Detection, so just pedestrian detection problem is converted into two classification problems, only need to distinguish pedestrian target and non-pedestrian target, and simplicity of design computing is fast.
Accompanying drawing explanation
Fig. 1-1st, one of HOG-CLBP feature extraction schematic diagram.
Fig. 1-2 is two of HOG-CLBP feature extraction schematic diagram.
Fig. 2 is NEU-person Sample Storehouse partial model figure.
Fig. 3 is system chart of the present invention.
Fig. 4 is the performance comparison result figure after the present invention and existing method are tested.
Fig. 5 is single goal testing result sample graph.
Fig. 6 is multi-target detection result sample graph.
Fig. 7 is artificial occlusion detection result sample graph from top to bottom.
Fig. 8 is artificially by left-to-right occlusion detection result sample graph.
Fig. 9 is that people is the testing result sample graph of blocking different proportion.
Figure 10 is actual object occlusion detection result sample graph.
Figure 11 is the pedestrian detection result sample graph under sleety weather disturbs.
Figure 12 is the pedestrian detection result sample graph under foggy weather disturbs.
Figure 13 is the pedestrian detection result sample graph adding under Gaussian noise condition.
Figure 14 is environment pedestrian detection result sample graph at night.
Figure 15 is the pedestrian detection result sample graph under camouflage condition.
Embodiment
Fig. 1-15 illustrate embodiments of the present invention by reference to the accompanying drawings;
Physical environment disturbs and has the pedestrian target detection method while blocking, and the method comprises the steps:
Step 1, set up pedestrian's training sample set, sample set is divided into the positive sample set that has pedestrian and the negative sample collection that there is no pedestrian; Sample is from NEU-Person Sample Storehouse and INRIA pedestrian's database;
Step 2, employing HOG-CLBP feature extracting method extract the characteristics of human body in sample set, by piecemeal, process image, the feature of extracting is carried out to PCA dimensionality reduction, obtain the global feature of pedestrian's training sample set sample;
Step 3, the feature of extraction is sent into svm classifier device classify, obtain the pedestrian's sorter based on NEU-Person Sample Storehouse;
Step 4, the pedestrian who extracts NEU-Person Sample Storehouse, calculate the approximate weighted value of all parts of pedestrian in NEU-person Sample Storehouse;
Step 5, complete laggard every trade people detection of training stage, with intensive scanning, detect testing image, by sliding window, operate, utilize the sorter training scan each window and adjudicate, the window of judging pedestrian carries out mark, finally the markd window of institute is merged;
Step 6, by image pyramid and multiple dimensioned sliding window, operate, the rectangle frame under the different yardstick of image is merged into a frame, and then obtains target position accurately, be about to pedestrian detection to be measured out.
HOG-CLBP feature extracting method in described step 2, adopt piecemeal to move the extraction of carrying out HOG-CLBP fusion feature, image in detection window is 64 * 128 pixels, to this image block, every block size is 16 * 16 pixels, by every, be divided into 4 unit again, each cell size is 8 * 8 pixels, as Figure 1-1; Add up the CLBP histogram of every, the every vectorial A that can obtain one 59 dimension; Then add up respectively the histogram of gradients of 4 unit in a piece, and merged, can obtain the vectorial B of one 36 dimension; Therefore a piece just can be used vectorial A, and B represents.
Fragmental image processing in step 2, wherein the moving step length of piece is the size of a unit, as shown in Figure 1-2, in a unit, from left to right, move from top to bottom, by AEFD, started, move to BGHC next time, move to again EIJF next time, until travel through whole detection window, finally will produce
(64/8-1) * (128/8-1)=105 piece, whole pictures can be with one
The vector representation of 105 * (59+36)=9975 dimensions.
In step 4, pedestrian's human body of NEU-Person Sample Storehouse is divided into 5 parts, 5 parts are respectively head, left arm and left half body portion, right arm and right half body portion, left and right thigh be a part and left and right shank and pin for a part of, occlusion issue can be effectively processed in division like this; By NEU-person Sample Storehouse, calculate the approximate weighted value of all parts, the weighted value of head is 0.2, and the weighted value of left arm and left half body is 0.29, and the weighted value of right arm and right half body is 0.29, the weighted value of two thighs in left and right is 0.15, and the weighted value of shank and pin is 0.07.Partial model as shown in Figure 2.
In step 6, build image pyramid:
With (n+2) first array (F
0, P
1..., P
n, the pedestrian target partial model that b) definition contains n subassembly; F wherein
0be used for representing whole wave filter (HOG-CLBP characteristic model), P
ibe the model of i subassembly, b is side-play amount; (the F of three-number set for partial model
i, v
i, d
i) represent, wherein, F
ibe the wave filter of i parts, size is w
i* h
i, v
ithe fixed position of the relative block mold of parts i, d
ibe the function that the relative fixed position of target component changes loss, subassembly departs from tram in its model; By the study of many group models, can obtain the different model of same target, and then be combined into partial model.
In step 6, target image is detected and is determined target by multiple dimensioned sliding window:
In target image, pedestrian's position is random and unfixed often, and pedestrian's size also can surpass the size of the standard detection window of 64*128, when detecting, first target image is carried out to intensive scanning, secondly target image is dwindled; Target image is dwindled by 3 layers or 4 layers, until the height of target image or width are less than or equal to height or the width of standard detection window, can obtain series of standards target image.
In the intensive testing process of the sliding window of described target image, can obtain a lot of rectangle frames; Yet these rectangle frames are not in same position, reason is that same pedestrian, under different yardsticks, has the skew of position with respect to former figure, and the rectangle frame producing just can not overlap; Can exert an influence to the definite of target final position; Therefore, these rectangle frames need to be merged into a frame, and obtain target position accurately.
In step 1, sample is from NEU-Person Sample Storehouse and INRIA pedestrian's database;
In described NEU-Person Sample Storehouse, training set has 613, positive sample, comprises 2362 pedestrians, relates to the environment such as rain, snow, mist, night and camouflage color; 227 of negative samples;
Test set has 187, positive sample, comprises 713 pedestrians, 150 of negative samples; Pedestrian's major part in Sample Storehouse is for stance and be highly greater than 100 pixels.
NEU-Person Sample Storehouse data set Shi You Northeastern University's Mechanical Academy Intelligent Measurement and quality controling research chamber institute external disclosure, in this Sample Storehouse, training set has 613, positive sample (to comprise 2362 pedestrians, relate to the environment such as rain, snow, mist, night and camouflage color), 227 of negative samples; Test set has 187, positive sample (comprising 713 pedestrians), 150 of negative samples.Pedestrian's major part in picture library is for stance and be highly greater than 100 pixels, and the sharpness of the source picture of image is higher, can normally read and show.
Experiment content and interpretation of result
Experiment one: application the present invention and existing CLBP method, the performance of HOG method and HOG-CLBP method contrasts, and experimental result is as shown in Figure 4.HOG-CLBP method component model of the present invention has beautiful and charming testing result as seen from the figure.
As can be seen from Figure 4,10
-3during FFPW, the loss based on LBP curve is very high, and reason is: LBP feature is strong to texture description ability, but edge information is insensitive.The result of the pedestrian detection based on HOG feature is 0.19, visible HOG feature has very strong advantage aspect pedestrian detection, reason is: HOG feature has very strong sign ability to the description of the marginal information of human body, and a large amount of background informations can be filtered when extracting HOG feature, be about to a large amount of invalid information and filter, and edge variation between human body and background is obvious.Have the resulting loss of HOG-CLBP curve of Fusion Features to be approximately 0.06, visible features fusion has very great help really to the raising of verification and measurement ratio, has gathered the advantage of HOG feature and CLBP feature.After adding partial model, the lower of loss reaches 0.02, and it is more accurate that visible parts model makes to detect, and proved the validity of the inventive method.
Experiment two: application the present invention under different physical environments and different pedestrian of blocking under condition detect.Wherein Fig. 5 is single goal testing result figure, Fig. 6 is multi-target detection result figure, Fig. 7 is for people is for blocking (from top to bottom) testing result figure, Fig. 8 is for people is for blocking (by left-to-right) testing result figure, Fig. 9 is for people is for blocking (different proportion) testing result figure, Figure 10 is actual object occlusion detection result figure, Figure 11 is the pedestrian detection result figure under sleety weather disturbs, Figure 12 is the pedestrian detection result figure under foggy weather disturbs, Figure 13 is the pedestrian detection result figure adding under Gaussian noise condition, Figure 14 is environment pedestrian detection result figure at night, Figure 15 is the pedestrian detection result figure under camouflage condition.
Experimental result shows that this method blocks the pedestrian of different proportion and type, and the detection of different visibility physical environments and the complex environments such as camouflage in various degree has good robustness.To adding in the detection of pedestrian in the picture of Gaussian noise of the same race not and night picture, also obtained good detection effect simultaneously.Therefore method in this paper is applicable to multiple different nature and people for blocking environment, has further proved the validity of the inventive method.
Claims (8)
1. physical environment disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
The method comprises the steps:
Step 1, set up pedestrian's training sample set, sample set is divided into the positive sample set that has pedestrian and the negative sample collection that there is no pedestrian; Sample is from NEU-Person Sample Storehouse and INRIA pedestrian's database;
Step 2, employing HOG-CLBP feature extracting method extract the characteristics of human body in sample set, by piecemeal, process image, the feature of extracting is carried out to PCA dimensionality reduction, obtain the global feature of pedestrian's training sample set sample;
Step 3, the feature of extraction is sent into svm classifier device classify, obtain the pedestrian's sorter based on NEU-Person Sample Storehouse;
Step 4, the pedestrian who extracts NEU-Person Sample Storehouse, calculate the approximate weighted value of all parts of pedestrian in NEU-person Sample Storehouse;
Step 5, complete laggard every trade people detection of training stage, with intensive scanning, detect testing image, by sliding window, operate, utilize the sorter training scan each window and adjudicate, the window of judging pedestrian carries out mark, finally the markd window of institute is merged;
Step 6, by image pyramid and multiple dimensioned sliding window, operate, the rectangle frame under the different yardstick of image is merged into a frame, and then obtains target position accurately, be about to pedestrian detection to be measured out.
2. a kind of physical environment according to claim 1 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
HOG-CLBP feature extracting method in described step 2, adopt piecemeal to move the extraction of carrying out HOG-CLBP fusion feature, image in detection window is 64 * 128 pixels, to this image block, every block size is 16 * 16 pixels, by every, be divided into 4 unit again, each cell size is 8 * 8 pixels; Add up the CLBP histogram of every, the every vectorial A that can obtain one 59 dimension; Then add up respectively the histogram of gradients of 4 unit in a piece, and merged, can obtain the vectorial B of one 36 dimension; Therefore a piece just can be used vectorial A, and B represents.
3. a kind of physical environment according to claim 1 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
Fragmental image processing in step 2, wherein the moving step length of piece is the size of a unit, in a unit from left to right, move from top to bottom, by AEFD, started, move to BGHC next time, move to again EIJF next time, until travel through whole detection window, finally will produce
(64/8-1) * (128/8-1)=105 piece, whole pictures can be with one
The vector representation of 105 * (59+36)=9975 dimensions.
4. a kind of physical environment according to claim 1 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
In step 4, pedestrian's human body of NEU-Person Sample Storehouse is divided into 5 parts, 5 parts are respectively head, left arm and left half body portion, right arm and right half body portion, left and right thigh be a part and left and right shank and pin for a part of, occlusion issue can be effectively processed in division like this; By NEU-person Sample Storehouse, calculate the approximate weighted value of all parts, the weighted value of head is 0.2, and the weighted value of left arm and left half body is 0.29, and the weighted value of right arm and right half body is 0.29, the weighted value of two thighs in left and right is 0.15, and the weighted value of shank and pin is 0.07.
5. a kind of physical environment according to claim 1 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
In step 6, build image pyramid:
With (n+2) first array (F
0, P
1..., P
n, the pedestrian target partial model that b) definition contains n subassembly; F wherein
0be used for representing whole wave filter, P
ibe the model of i subassembly, b is side-play amount; (the F of three-number set for partial model
i, v
i, d
i) represent, wherein, F
ibe the wave filter of i parts, size is w
i* h
i, v
ithe fixed position of the relative block mold of parts i, d
iit is the function that the relative fixed position of target component changes loss; By the study of many group models, can obtain the different model of same target, and then be combined into partial model.
6. a kind of physical environment according to claim 1 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
In step 6, target image is detected and is determined target by multiple dimensioned sliding window:
In target image, pedestrian's position is random and unfixed often, and pedestrian's size also can surpass the size of the standard detection window of 64*128, when detecting, first target image is carried out to intensive scanning, secondly target image is dwindled; Target image is dwindled by 3 layers or 4 layers, until the height of target image or width are less than or equal to height or the width of standard detection window, can obtain series of standards target image.
7. a kind of physical environment according to claim 6 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
In the intensive testing process of the sliding window of described target image, can obtain a lot of rectangle frames; Yet these rectangle frames are not in same position, reason is that same pedestrian, under different yardsticks, has the skew of position with respect to former figure, and the rectangle frame producing just can not overlap; Can exert an influence to the definite of target final position; Therefore, these rectangle frames need to be merged into a frame, and obtain target position accurately.
8. a kind of physical environment according to claim 6 disturbs and has the pedestrian target detection method while blocking, and it is characterized in that:
In step 1, sample is from NEU-Person Sample Storehouse and INRIA pedestrian's database;
In described NEU-Person Sample Storehouse, training set has 613, positive sample, comprises 2362 pedestrians, relates to the environment such as rain, snow, mist, night and camouflage color; 227 of negative samples;
Test set has 187, positive sample, comprises 713 pedestrians, 150 of negative samples; Pedestrian's major part in Sample Storehouse is for stance and be highly greater than 100 pixels.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310643883.3A CN103617426B (en) | 2013-12-04 | 2013-12-04 | Pedestrian target detection method under interference by natural environment and shelter |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310643883.3A CN103617426B (en) | 2013-12-04 | 2013-12-04 | Pedestrian target detection method under interference by natural environment and shelter |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103617426A true CN103617426A (en) | 2014-03-05 |
CN103617426B CN103617426B (en) | 2017-02-01 |
Family
ID=50168129
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310643883.3A Active CN103617426B (en) | 2013-12-04 | 2013-12-04 | Pedestrian target detection method under interference by natural environment and shelter |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103617426B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105138983A (en) * | 2015-08-21 | 2015-12-09 | 燕山大学 | Pedestrian detection method based on weighted part model and selective search segmentation |
CN105335716A (en) * | 2015-10-29 | 2016-02-17 | 北京工业大学 | Improved UDN joint-feature extraction-based pedestrian detection method |
CN105590116A (en) * | 2015-12-18 | 2016-05-18 | 华南理工大学 | Bird image identification method based on head part alignment |
CN106067041A (en) * | 2016-06-03 | 2016-11-02 | 河海大学 | A kind of multi-target detection method of based on rarefaction representation of improvement |
CN106485283A (en) * | 2016-10-19 | 2017-03-08 | 电子科技大学 | A kind of particle filter pedestrian target tracking based on Online Boosting |
CN106611156A (en) * | 2016-11-03 | 2017-05-03 | 桂林电子科技大学 | Pedestrian recognition method and system capable of self-adapting to deep space features |
CN107273799A (en) * | 2017-05-11 | 2017-10-20 | 上海斐讯数据通信技术有限公司 | A kind of indoor orientation method and alignment system |
CN107463895A (en) * | 2017-07-28 | 2017-12-12 | 中国科学院西安光学精密机械研究所 | Small and weak damage object detection method based on neighborhood vector PCA |
CN107766848A (en) * | 2017-11-24 | 2018-03-06 | 广州鹰瞰信息科技有限公司 | The pedestrian detection method and storage medium of vehicle front |
CN107844765A (en) * | 2017-10-31 | 2018-03-27 | 广东欧珀移动通信有限公司 | Photographic method, device, terminal and storage medium |
CN107886066A (en) * | 2017-11-07 | 2018-04-06 | 中国矿业大学 | A kind of pedestrian detection method based on improvement HOG SSLBP |
CN108154099A (en) * | 2017-12-20 | 2018-06-12 | 北京奇艺世纪科技有限公司 | A kind of character recognition method, device and electronic equipment |
CN108197543A (en) * | 2017-12-22 | 2018-06-22 | 深圳云天励飞技术有限公司 | Image filtering method and device, embedded device and storage medium |
CN108197618A (en) * | 2018-04-08 | 2018-06-22 | 百度在线网络技术(北京)有限公司 | For generating the method and apparatus of Face datection model |
CN108846843A (en) * | 2018-04-03 | 2018-11-20 | 南昌奇眸科技有限公司 | A kind of image characteristic extracting method |
CN113538455A (en) * | 2021-06-15 | 2021-10-22 | 聚好看科技股份有限公司 | Three-dimensional hairstyle matching method and electronic equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080050041A1 (en) * | 2003-05-23 | 2008-02-28 | Lockhead Martin Corporation | Method and apparatus for filtering an image |
CN101887524A (en) * | 2010-07-06 | 2010-11-17 | 湖南创合制造有限公司 | Pedestrian detection method based on video monitoring |
US20120076408A1 (en) * | 2010-09-29 | 2012-03-29 | Andong University Industry-Academic Cooperation Foundation | Method and system for detecting object |
CN102609720A (en) * | 2012-01-31 | 2012-07-25 | 中国科学院自动化研究所 | Pedestrian detection method based on position correction model |
-
2013
- 2013-12-04 CN CN201310643883.3A patent/CN103617426B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080050041A1 (en) * | 2003-05-23 | 2008-02-28 | Lockhead Martin Corporation | Method and apparatus for filtering an image |
CN101887524A (en) * | 2010-07-06 | 2010-11-17 | 湖南创合制造有限公司 | Pedestrian detection method based on video monitoring |
US20120076408A1 (en) * | 2010-09-29 | 2012-03-29 | Andong University Industry-Academic Cooperation Foundation | Method and system for detecting object |
CN102609720A (en) * | 2012-01-31 | 2012-07-25 | 中国科学院自动化研究所 | Pedestrian detection method based on position correction model |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105138983A (en) * | 2015-08-21 | 2015-12-09 | 燕山大学 | Pedestrian detection method based on weighted part model and selective search segmentation |
CN105138983B (en) * | 2015-08-21 | 2019-06-28 | 燕山大学 | The pedestrian detection method divided based on weighting block model and selective search |
CN105335716B (en) * | 2015-10-29 | 2019-03-26 | 北京工业大学 | A kind of pedestrian detection method extracting union feature based on improvement UDN |
CN105335716A (en) * | 2015-10-29 | 2016-02-17 | 北京工业大学 | Improved UDN joint-feature extraction-based pedestrian detection method |
CN105590116A (en) * | 2015-12-18 | 2016-05-18 | 华南理工大学 | Bird image identification method based on head part alignment |
CN105590116B (en) * | 2015-12-18 | 2019-05-14 | 华南理工大学 | A kind of birds image-recognizing method based on head piece alignment |
CN106067041A (en) * | 2016-06-03 | 2016-11-02 | 河海大学 | A kind of multi-target detection method of based on rarefaction representation of improvement |
CN106067041B (en) * | 2016-06-03 | 2019-05-31 | 河海大学 | A kind of improved multi-target detection method based on rarefaction representation |
CN106485283A (en) * | 2016-10-19 | 2017-03-08 | 电子科技大学 | A kind of particle filter pedestrian target tracking based on Online Boosting |
CN106611156A (en) * | 2016-11-03 | 2017-05-03 | 桂林电子科技大学 | Pedestrian recognition method and system capable of self-adapting to deep space features |
CN106611156B (en) * | 2016-11-03 | 2019-12-20 | 桂林电子科技大学 | Pedestrian identification method and system based on self-adaptive depth space characteristics |
CN107273799A (en) * | 2017-05-11 | 2017-10-20 | 上海斐讯数据通信技术有限公司 | A kind of indoor orientation method and alignment system |
CN107463895B (en) * | 2017-07-28 | 2019-09-20 | 中国科学院西安光学精密机械研究所 | Small and weak damage object detection method based on neighborhood vector PCA |
CN107463895A (en) * | 2017-07-28 | 2017-12-12 | 中国科学院西安光学精密机械研究所 | Small and weak damage object detection method based on neighborhood vector PCA |
CN107844765A (en) * | 2017-10-31 | 2018-03-27 | 广东欧珀移动通信有限公司 | Photographic method, device, terminal and storage medium |
CN107886066A (en) * | 2017-11-07 | 2018-04-06 | 中国矿业大学 | A kind of pedestrian detection method based on improvement HOG SSLBP |
CN107766848A (en) * | 2017-11-24 | 2018-03-06 | 广州鹰瞰信息科技有限公司 | The pedestrian detection method and storage medium of vehicle front |
CN108154099A (en) * | 2017-12-20 | 2018-06-12 | 北京奇艺世纪科技有限公司 | A kind of character recognition method, device and electronic equipment |
CN108154099B (en) * | 2017-12-20 | 2021-04-30 | 北京奇艺世纪科技有限公司 | Figure identification method and device and electronic equipment |
CN108197543A (en) * | 2017-12-22 | 2018-06-22 | 深圳云天励飞技术有限公司 | Image filtering method and device, embedded device and storage medium |
CN108846843A (en) * | 2018-04-03 | 2018-11-20 | 南昌奇眸科技有限公司 | A kind of image characteristic extracting method |
CN108197618A (en) * | 2018-04-08 | 2018-06-22 | 百度在线网络技术(北京)有限公司 | For generating the method and apparatus of Face datection model |
CN113538455A (en) * | 2021-06-15 | 2021-10-22 | 聚好看科技股份有限公司 | Three-dimensional hairstyle matching method and electronic equipment |
CN113538455B (en) * | 2021-06-15 | 2023-12-12 | 聚好看科技股份有限公司 | Three-dimensional hairstyle matching method and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN103617426B (en) | 2017-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103617426A (en) | Pedestrian target detection method under interference by natural environment and shelter | |
CN103473571B (en) | Human detection method | |
CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
CN103442209B (en) | Video monitoring method of electric transmission line | |
CN105260712B (en) | A kind of vehicle front pedestrian detection method and system | |
CN103902976B (en) | A kind of pedestrian detection method based on infrared image | |
CN104036239B (en) | Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering | |
CN100545867C (en) | Aerial shooting traffic video frequency vehicle rapid checking method | |
CN102609720B (en) | Pedestrian detection method based on position correction model | |
CN104091171B (en) | Vehicle-mounted far infrared pedestrian detecting system and method based on local feature | |
CN102663411B (en) | Recognition method for target human body | |
CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
CN103310194A (en) | Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction | |
CN105404857A (en) | Infrared-based night intelligent vehicle front pedestrian detection method | |
CN103390164A (en) | Object detection method based on depth image and implementing device thereof | |
CN109902618A (en) | A kind of sea ship recognition methods and device | |
Xu et al. | Real-time pedestrian detection based on edge factor and Histogram of Oriented Gradient | |
CN103778435A (en) | Pedestrian fast detection method based on videos | |
Xia et al. | A novel algorithm for ship detection based on dynamic fusion model of multi-feature and support vector machine | |
CN103745197B (en) | A kind of detection method of license plate and device | |
CN106204594A (en) | A kind of direction detection method of dispersivity moving object based on video image | |
CN109886086B (en) | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier | |
CN104680154B (en) | A kind of personal identification method merged based on face characteristic and palm print characteristics | |
CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |