CN107038416B - Pedestrian detection method based on binary image improved HOG characteristics - Google Patents

Pedestrian detection method based on binary image improved HOG characteristics Download PDF

Info

Publication number
CN107038416B
CN107038416B CN201710140483.9A CN201710140483A CN107038416B CN 107038416 B CN107038416 B CN 107038416B CN 201710140483 A CN201710140483 A CN 201710140483A CN 107038416 B CN107038416 B CN 107038416B
Authority
CN
China
Prior art keywords
image
binary image
gradient
pixel
pedestrian detection
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
Application number
CN201710140483.9A
Other languages
Chinese (zh)
Other versions
CN107038416A (en
Inventor
冯颖
杨涛
苏比哈什·如凯迦
陈新开
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201710140483.9A priority Critical patent/CN107038416B/en
Publication of CN107038416A publication Critical patent/CN107038416A/en
Application granted granted Critical
Publication of CN107038416B publication Critical patent/CN107038416B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a pedestrian detection method based on binary image improved HOG characteristics, which comprises the following steps: establishing a pedestrian training sample library; carrying out binarization processing on the sample image; extracting an improved HOG characteristic vector from the binary image; carrying out Gaussian normalization on the improved HOG characteristic vector, training by utilizing positive and negative samples to obtain each parameter in the SVM model, and establishing a linear SVM model; preprocessing a video frame image to be detected to obtain a binary image; calculating to obtain a current improved HOG characteristic vector; inputting the improved HOG feature vector into a linear SVM model, if the output of the model is judged to be a positive sample, detecting a target, and outputting the position of the target; and detecting whether each traversed window has a target or not in a window traversal mode. The pedestrian detection method and the pedestrian detection device can effectively overcome the defects of high memory consumption and low detection speed in the pedestrian detection process.

Description

Pedestrian detection method based on binary image improved HOG characteristics
Technical Field
The invention relates to the field of computer vision pedestrian detection, in particular to a pedestrian detection method based on binary image improved HOG characteristics.
Background
At present, pedestrian detection methods are mainly classified into two categories: template matching based methods and statistical learning based methods. The pedestrian detection algorithm based on statistical learning is gradually widely applied because the various performances of the statistical algorithm meet various application occasions.
In the statistical learning-based method, a Histogram of Oriented Gradients (HOG) feature is a very effective gradient feature, which has good robustness for image pedestrian detection under illumination variation and color variation. The method comprises the steps of firstly dividing an image into small connected regions, defining the connected regions as cell units, collecting direction histograms of gradients or edges of all pixel points in the cell units, and then combining the histograms to form the feature descriptor. After obtaining the local histogram, the local histogram of the cell unit statistics needs to be contrast normalized over a larger range of the image. Typically, local histogram features within a block consisting of four cell units are computed, and then 4 cell units within the block are normalized separately. The normalized features allow the detected image to be more robust to illumination variations and shadows. And cascading the four normalized local histogram features to obtain an HOG feature vector. The HOG feature has a good effect in shape-based object detection such as pedestrian detection, and many subsequent target detection algorithms are extensions on the basis.
Some problems still exist in pedestrian detection in the field of intelligent security at the present stage: in the field of real-time detection, the real-time requirement is very high, and the real-time performance is poor due to the increase of the feature detection computation amount; in video pedestrian detection, a pyramid model of a gray level image needs to occupy a large memory resource. Therefore, a pedestrian detection method overcoming the problems is sought, and has important research significance and practical value.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a pedestrian detection method based on the binary image improved HOG characteristics, and can effectively overcome the defects of high memory consumption and low detection speed in the pedestrian detection process.
The purpose of the invention is realized by the following technical scheme: a pedestrian detection method based on binary image improved HOG characteristics comprises the following steps:
a learning stage:
s1, establishing a pedestrian training sample library;
s2, performing binarization processing on the sample image by using a line block pixel-based local adaptive binarization method to obtain a binarized image;
s3, extracting an improved HOG characteristic vector from the binary image;
s4, carrying out Gaussian normalization on the improved HOG feature vector, training by using positive and negative samples to obtain each parameter in the SVM model, and establishing a linear SVM model;
a decision stage:
s5, preprocessing a video frame image to be detected to obtain a binary image;
s6, calculating to obtain the current improved HOG characteristic vector;
s7, inputting the improved HOG feature vector into the linear SVM model obtained in the step S4, if the output of the model is judged to be a positive sample, detecting a target, and outputting the position of the target; and detecting whether each traversed window has a target or not in a window traversal mode.
Preferably, the selection of the pedestrian training sample library in step S1 follows the following two rules:
rule 1: the number ratio of the pedestrian negative samples to the pedestrian positive samples is 10: 1;
rule 2: by utilizing the SVM trained for the first time, the negative samples which are detected by mistake in the negative samples are taken as difficult negative samples, the proportion of the difficult negative samples is improved, and therefore the accuracy of the established model can be further improved.
Preferably, the local adaptive binarization method based on line block pixels in step S2 specifically includes: when reading video data in the FPGA, only two lines of image stream information need to be stored in the RAM; updating the gray value of the pixel of the ith row by using the gray value of the gray map of the ith row and the gray value of the updated row vector image of the (i-1) th row, wherein the updated value of the pixel of the 1 st row is the original gray value per se, and the formula is as follows:
wherein preY (i, j) refers to the updated row vector image gray value, cur (i, j) refers to the gray value of the gray image;
the binarization threshold value of each pixel point is determined as follows: taking a pixel area with the size of 1 x w line block for the pixels corresponding to the updated line vector, obtaining a mean value E (i, j) of the area, determining a binarization threshold value of the obtained pixel points according to the mean value E (i, j) of the pixel points in the line block, and obtaining a binarization image by using the following formula:
Figure GDA0002236670680000032
where I (I, j) is the grayscale value of the binarized image, and σ is the threshold coefficient.
Preferably, the specific steps of step S3 are: and extracting the gradient amplitude and angle of each pixel point by using a gradient template, and performing histogram statistics according to the gradient direction to obtain an improved HOG feature vector.
Specifically, the gradient information of the local area of the binary image is calculated by using a 5-by-5 gradient template, when the template calculation is carried out, a central pixel point is corresponding to h (0,0), and then pixel points around the central pixel point are multiplied by pixel points h (a, b) corresponding to the template; the following formula is a general method for calculating the gradient:
Figure GDA0002236670680000033
wherein f is a binary image, and (x, y) are pixel points, and for a template of 5 × 5, the values of a and b can be +/-2, +/-1 and 0;
by using hxCalculating a direction gradient template to obtain a gradient g in the x directionx(x, y) by hyCalculating a y-direction gradient g by a direction gradient templatey(x, y); the formula for calculating the gradient amplitude and angle is respectively as follows:
Figure GDA0002236670680000034
and then carrying out histogram statistics according to the gradient amplitude of each pixel point and the gradient direction to obtain the improved HOG characteristic vector.
Further, a direction interval unequal division mode is adopted when histogram statistics is carried out, and 0-pi intervals are divided according to the number of the direction intervals
Figure GDA0002236670680000045
The interval is divided into an interval,
Figure GDA0002236670680000042
Interval is according to
Figure GDA0002236670680000043
Is divided into 5 intervals at equal intervals、
Figure GDA0002236670680000044
The histogram is divided into 7 bins, so that the histogram has 7 channels.
Specifically, an image detection window size 64 × 128 is selected, a block size is 16 × 16, a cell size is 8 × 8, a block offset is 8 × 8, and the number of blocks is 105, each cell includes 8 × 8 pixels, and binary pixel gradient information of 8 × 8 pixels is voted by using different direction angles; the histogram voting adopts weighted voting, namely the gradient amplitude of each pixel is taken as voting weight to obtain a multidimensional characteristic vector.
Preferably, in step S4, gaussian normalization is performed on each element in the feature vector, so that 99% of the obtained component values fall within [0, 1%]Obtaining a normalized improved HOG characteristic vector; inputting the normalized improved HOG feature vector into a support vector machine classification model for training so as to obtain a pedestrian detection model of the improved HOG feature; establishing a linear SVM model: f (x) wTx + b, x is the normalized modified HOG feature vector, wTAnd b is the SVM parameter obtained by positive and negative sample training.
Preferably, the preprocessing in step S5 includes the following specific steps: extracting a frame image from an input video, converting an RGB image into a gray map, filtering, and then performing binarization processing on a sample image by using a line block pixel-based local adaptive binarization method to obtain a binarized image.
Preferably, in step S6, a pyramid model is established for the binarized image of the region of interest, level0 refers to the binarized image, level1 refers to the image obtained by reducing the binarized image by 0.95 times, and so on until the resolution of the image is not greater than 64 × 64, and the scaling rate of each layer is 0.95; and calculating the improved HOG characteristic vector of each layer of image in sequence and carrying out Gaussian normalization processing to obtain the normalized improved HOG characteristic vector.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the invention utilizes the binary image to detect the pedestrian, and can greatly improve the real-time performance of the detection system and reduce the memory consumption compared with an 8-bit gray-scale image.
2. The invention reduces the multi-line image data stored in the memory to two-line image data by using a line block pixel-based local self-adaptive binarization method, thereby greatly reducing the storage space consumption.
3. The improved HOG characteristic vector is used as the characteristic vector of the binary image, the HOG characteristic angle information of the binary image is greatly reserved, a large amount of effective gradient information is reserved for subsequent histogram statistics, and the pedestrian detection accuracy is improved.
Drawings
FIG. 1 is a flowchart of the method of the present embodiment;
FIG. 2 is a flow chart of an improved HOG feature extraction for binary images;
FIG. 3 is a diagram of a gradient template: FIG. 3(a) is an X-direction gradient template; FIG. 3(b) is a Y-direction gradient template;
fig. 4 is a division diagram of the improved HOG feature direction section.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
A pedestrian detection method based on binary image improved HOG characteristics, as shown in figure 1, comprises the following steps:
a learning stage:
and S1, establishing a pedestrian training sample library.
The selection of the pedestrian training sample library follows the following two rules:
rule 1: the number ratio of the pedestrian negative samples to the pedestrian positive samples is 10: 1;
rule 2: by utilizing the SVM trained for the first time, the negative samples which are detected by mistake in the negative samples are taken as difficult negative samples, the proportion of the difficult negative samples is improved, and therefore the accuracy of the established model can be further improved.
And S2, carrying out binarization processing on the sample image.
When the traditional local adaptive binarization method is applied to FPGA, the storage space of a plurality of lines (n lines) of image streams needs to be reserved in the input line image streams. Instead of directly retaining n rows of gray image pixels, the present embodiment updates the updated value of the ith row of pixels by using the gray image gray value of the ith row together with the updated gray image gray value of the (i-1) th row of vector image (the updated value of the 1 st row of pixels is the original gray value itself), as follows:
where preY (i, j) refers to the updated row vector image grayscale value, cur (i, j) refers to the grayscale image grayscale value. When reading video data, only two lines of image stream information need to be saved in the RAM, which is referred to as a line block pixel-based local adaptive binarization method in this embodiment. Carrying out binarization processing on the sample image by using a local self-adaptive binarization method based on row block pixels, wherein a binarization threshold value of each pixel point is determined as follows: taking a pixel area with the size of 1 x w line block for the pixels corresponding to the updated line vector, obtaining a mean value E (i, j) of the area, determining a binarization threshold value of the obtained pixel points according to the mean value E (i, j) of the pixel points in the line block, and obtaining a binarization image by using the following formula:
where I (I, j) is the grayscale value of the binarized image, and σ is the threshold coefficient.
And S3, extracting the improved HOG characteristic of the binary image, as shown in figure 2.
And extracting the gradient amplitude and angle of each pixel point by using a gradient template, and performing histogram statistics according to the gradient direction to obtain an improved HOG feature vector. And gradient amplitude and angle are calculated, gradient calculation is completed by utilizing a gradient template, and edge gradient characteristics are very sensitive to the gradient template.
And calculating the gradient information of the local area of the binary image by using a 5-by-5 gradient template, and counting a gradient direction histogram to form an improved HOG feature. Firstly, the gradient amplitude and the angle of the traditional first-order gradient template are calculated by using the 5-by-5 gradient template instead of the traditional first-order gradient template, the gradient amplitude direction of the binary image calculated by the traditional first-order gradient template is only four angle directions, and the gradient direction generated by using the 5-by-5 gradient template can keep the gradient direction information, so that the angle characteristics of the HOG characteristic vector are greatly enriched, and the phenomenon of low detection rate caused by over-sparse angle directions is avoided.
Here, as shown in fig. 3, when calculating the gradient template of 5 × 5, the central pixel point is associated with h (0,0), and then the pixel points around the central pixel point are multiplied by the pixel points h (a, b) associated with the template. The following formula is a general method for calculating the gradient:
Figure GDA0002236670680000063
wherein f is the binary image, and (x, y) are the pixel points, and for a template of 5 × 5, the values of a and b can be ± 2, ± 1 and 0.
By using hxCalculating a direction gradient template to obtain a gradient g in the x directionx(x, y) by hyCalculating a y-direction gradient g by a direction gradient templatey(x, y). The formula for calculating the gradient amplitude and angle is respectively as follows:
Figure GDA0002236670680000072
then, histogram statistics is carried out according to the gradient amplitude of each pixel point and the gradient direction, and because the gradient angle of the HOG characteristic of the binary image is concentrated in the angle intervals of 0 and pi, a direction interval unequal division mode is adopted, for example, as shown in FIG. 4, the interval of 0-pi is divided according to the gradient direction
Figure GDA0002236670680000077
The interval is divided into an interval,
Figure GDA0002236670680000074
Interval is according to
Figure GDA0002236670680000075
Is divided into 5 intervals at equal intervals,
Figure GDA0002236670680000076
The histogram is divided into 7 bins (bins), so that the histogram has 7 channels, and the statistical result of the histogram is the improved HOG feature vector.
An image detection window size of 64 × 128, a block size of 16 × 16, a cell size of 8 × 8, a block offset of 8 × 8, and a block number of 105 were selected, each cell including 8 × 8 pixels, and binary pixel gradient information of 8 × 8 pixels was voted using 7 direction angles. Specifically, the histogram voting adopts weighted voting, i.e. the gradient magnitude of each pixel is used as the voting weight, resulting in a 7-dimensional feature vector. Each detection box has 2940-dimensional improved HOG feature vector.
And S4, carrying out Gaussian normalization on the improved HOG feature vector, training by using positive and negative samples to obtain each parameter in the SVM model, and establishing the linear SVM model.
And performing Gaussian normalization processing on each element in the feature vector to enable 99% of the value range of the obtained component value to fall between [0 and 1], so as to obtain the normalized improved HOG feature vector. And inputting the normalized improved HOG feature vector into a classification model of a support vector machine for training, thereby obtaining a pedestrian detection model of the improved HOG feature.
Establishing a linear SVM model: f (x) wTx + b, x is the normalized modified HOG feature vector, which is also the edge gradient feature, wTAnd b is the SVM parameter obtained by positive and negative sample training.
A decision stage:
and S5, preprocessing the video frame image to be detected to obtain a binary image.
In this embodiment, in the field of pedestrian detection, the preprocessing includes the specific steps of: extracting a frame image from an input video, converting an RGB image into a gray map, filtering, and performing binarization processing on a sample image based on line block pixel local self-adaptive binarization to obtain a binarized image.
And S6, calculating to obtain the current improved HOG feature vector.
Different from scaling detection of an 8-bit gray level image in traditional video detection, in this embodiment, an input video frame image is binarized to obtain a 1-bit binarized image, and a pyramid scaling model of the binarized image is established. The most memory consumption and the most time consumption in pedestrian detection are the pyramid scaling and detection of the image, and the 8-bit gray image is converted into a 1-bit binary image, so that the memory consumption is greatly reduced, and the detection speed is accelerated.
And establishing a pyramid model for the binary image, wherein a level0 layer refers to the binary image, a level1 layer refers to the image obtained by reducing the binary image by 0.95 time, and the like until the resolution of the image is not more than 64 × 64, and the scaling rate of each layer is 0.95.
And calculating the improved HOG characteristic vector of each layer of image in sequence and carrying out Gaussian normalization processing to obtain the normalized improved HOG characteristic vector.
And S7, inputting the linear SVM model obtained in the step S4, and if the model output is judged to be a positive sample, detecting the target and outputting the target position.
According to f (x) w in the linear SVM modelTAnd (4) introducing the normalized improved HOG characteristic vector into the model parameters obtained by x + b learning, and outputting the target position according to the rule that more than 0 is a positive sample.
And for the region of interest, detecting whether each traversed window has a target or not in a window traversal mode by using a pyramid model.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (7)

1. A pedestrian detection method based on binary image improved HOG features is characterized by comprising the following steps:
a learning stage:
s1, establishing a pedestrian training sample library;
s2, carrying out binarization processing on the sample image by using a line block pixel-based local adaptive binarization method to obtain a binary image;
the local adaptive binarization method based on the line block pixels in the step S2 specifically includes: when reading video data in the FPGA, only two lines of image stream information need to be stored in the RAM; updating the gray value of the pixel of the ith row by using the gray value of the gray map of the ith row and the gray value of the updated row vector image of the (i-1) th row, wherein the updated value of the pixel of the 1 st row is the original gray value per se, and the formula is as follows:
wherein preY (i, j) refers to the updated row vector image gray value, cur (i, j) refers to the gray value of the gray image;
the binarization threshold value of each pixel point is determined as follows: taking a pixel region with the size of 1 x w line block for the pixels corresponding to the updated line vector, obtaining a mean value E (i, j) of the region, determining a binary threshold value of the obtained pixel points according to the mean value E (i, j) of the pixel points in the line block, and obtaining a binary image by using the following formula:
wherein I (I, j) is the gray value of the binary image, and sigma is the threshold coefficient;
s3, extracting an improved HOG feature vector from the binary image;
calculating gradient information of a binary image local region by using a 5 x 5 gradient template, corresponding a central pixel point of the binary image local region to a pixel point h (0,0) corresponding to the gradient template when performing template calculation, and multiplying pixel points around the central pixel point of the binary image local region by pixel points h (a, b) corresponding to the gradient template; assuming that f (x, y) is equal to I (I, j), the gradient of the local region of the binary image is calculated by the following formula:
Figure FDA0002260257260000013
wherein f is a binary image, and (x, y) are pixel coordinates, and for a template of 5 × 5, the values of a and b can be ± 2, ± 1 and 0;
by using hxCalculating a direction gradient template to obtain a gradient g in the x directionx(x, y) by hyCalculating a y-direction gradient g by a direction gradient templatey(x, y); the formula for calculating the gradient amplitude and angle is respectively as follows:
Figure FDA0002260257260000021
carrying out histogram statistics according to the gradient amplitude of each pixel point and the gradient direction to obtain an improved HOG characteristic vector;
s4, carrying out Gaussian normalization on the improved HOG feature vector, training by using positive and negative samples to obtain each parameter in the SVM model, and establishing a linear SVM model;
a decision stage:
s5, preprocessing a video frame image to be detected to obtain a binary image;
s6, calculating to obtain the current improved HOG characteristic vector;
s7, inputting the improved HOG feature vector into the linear SVM model obtained in the step S4, if the output of the model is judged to be a positive sample, detecting a target, and outputting the position of the target; and detecting whether each traversed window has a target or not in a window traversal mode.
2. The pedestrian detection method according to claim 1, wherein the selection of the pedestrian training sample library in step S1 follows the following two rules:
rule 1: the number ratio of the pedestrian negative samples to the pedestrian positive samples is 10: 1;
rule 2: by utilizing the SVM trained for the first time, the negative samples which are detected by mistake in the negative samples are taken as difficult negative samples, the proportion of the difficult negative samples is improved, and therefore the accuracy of the established model can be further improved.
3. The pedestrian detection method according to claim 1, wherein the histogram statistics is performed by adopting a direction interval unequal manner, and the interval of 0 to pi is divided into
Figure FDA0002260257260000023
The interval is divided into an interval,
Figure FDA0002260257260000024
Interval is according to
Figure FDA0002260257260000025
Is divided into 5 intervals at equal intervals,The histogram is divided into 7 intervals, and the histogram has 7 channels.
4. The pedestrian detection method according to claim 1, wherein an image detection window size 64 x 128, a block size 16 x 16, a cell size 8 x 8, a block offset 8 x 8, a block number 105, each cell comprising 8 x 8 pixels, is selected, and binary pixel gradient information of 8 x 8 pixels is voted using different directional angles; the histogram voting adopts weighted voting, namely the gradient amplitude of each pixel is taken as voting weight to obtain a multidimensional characteristic vector.
5. The pedestrian detection method according to claim 1, wherein in step S4, each element in the improved HOG feature vector is subjected to gaussian normalization processing so that 99% of the obtained component values fall within a range of 99%[0,1]Obtaining a normalized improved HOG characteristic vector; inputting the normalized improved HOG feature vector into a support vector machine classification model for training so as to obtain a pedestrian detection model of the improved HOG feature; establishing a linear SVM model: f (x) wTx + b, x is the normalized modified HOG feature vector, wTAnd b is the SVM parameter obtained by positive and negative sample training.
6. The pedestrian detection method according to claim 1, wherein the preprocessing in step S5 includes the specific steps of: extracting a frame image from an input video, converting an RGB image into a gray map, filtering, and then performing binarization processing on a sample image by using a line block pixel-based local adaptive binarization method to obtain a binary image.
7. The pedestrian detection method according to claim 1, wherein in step S6, a pyramid model is built for the region-of-interest binary image, level0 refers to the region-of-interest binary image, level1 refers to the image obtained by reducing the region-of-interest binary image by 0.95 times, and so on until the resolution of the region-of-interest binary image is not greater than 64 × 64, and the scaling rate of each layer is 0.95; and sequentially calculating an improved HOG characteristic vector for each layer of the two-value image of the region of interest and carrying out Gaussian normalization processing to obtain a normalized improved HOG characteristic vector.
CN201710140483.9A 2017-03-10 2017-03-10 Pedestrian detection method based on binary image improved HOG characteristics Active CN107038416B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710140483.9A CN107038416B (en) 2017-03-10 2017-03-10 Pedestrian detection method based on binary image improved HOG characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710140483.9A CN107038416B (en) 2017-03-10 2017-03-10 Pedestrian detection method based on binary image improved HOG characteristics

Publications (2)

Publication Number Publication Date
CN107038416A CN107038416A (en) 2017-08-11
CN107038416B true CN107038416B (en) 2020-02-18

Family

ID=59534454

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710140483.9A Active CN107038416B (en) 2017-03-10 2017-03-10 Pedestrian detection method based on binary image improved HOG characteristics

Country Status (1)

Country Link
CN (1) CN107038416B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107679528A (en) * 2017-11-24 2018-02-09 广西师范大学 A kind of pedestrian detection method based on AdaBoost SVM Ensemble Learning Algorithms
CN108182442B (en) * 2017-12-29 2022-03-15 惠州华阳通用电子有限公司 Image feature extraction method
CN108460420A (en) * 2018-03-13 2018-08-28 江苏实达迪美数据处理有限公司 A method of classify to certificate image
CN108710909B (en) * 2018-05-17 2023-01-31 南京汇川工业视觉技术开发有限公司 Counting method for deformable, rotary and invariant boxed objects
CN108875628A (en) * 2018-06-14 2018-11-23 攀枝花学院 pedestrian detection method
CN109034125B (en) * 2018-08-30 2021-12-03 北京工业大学 Pedestrian detection method and system based on scene complexity
CN110866534B (en) * 2019-08-13 2023-09-12 广州三木智能科技有限公司 Far infrared pedestrian training method for gradient amplitude distribution gradient orientation histogram
CN110765877B (en) * 2019-09-20 2022-09-06 南京理工大学 Pedestrian detection method and system based on thermal imager and binocular camera
CN111913584B (en) * 2020-08-19 2022-04-01 福州大学 Mouse cursor control method and system based on gesture recognition

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663409B (en) * 2012-02-28 2015-04-22 西安电子科技大学 Pedestrian tracking method based on HOG-LBP
CN103177262B (en) * 2013-02-19 2016-01-20 山东大学 Based on FPGA framework and its implementation of the pedestrian detecting system of HOG and SVM
CN103440035A (en) * 2013-08-20 2013-12-11 华南理工大学 Gesture recognition system in three-dimensional space and recognition method thereof
CN103530638B (en) * 2013-10-29 2016-08-17 无锡赛思汇智科技有限公司 Method for pedestrian matching under multi-cam
CN103745226B (en) * 2013-12-31 2015-03-18 国家电网公司 Dressing safety detection method for worker on working site of electric power facility

Also Published As

Publication number Publication date
CN107038416A (en) 2017-08-11

Similar Documents

Publication Publication Date Title
CN107038416B (en) Pedestrian detection method based on binary image improved HOG characteristics
CN108121991B (en) Deep learning ship target detection method based on edge candidate region extraction
CN109840521B (en) Integrated license plate recognition method based on deep learning
CN104834912B (en) A kind of weather recognition methods and device based on image information detection
CN107507173B (en) No-reference definition evaluation method and system for full-slice image
CN110334762B (en) Feature matching method based on quad tree combined with ORB and SIFT
CN109840483B (en) Landslide crack detection and identification method and device
CN107506765B (en) License plate inclination correction method based on neural network
CN107871316B (en) Automatic X-ray film hand bone interest area extraction method based on deep neural network
CN111127360B (en) Gray image transfer learning method based on automatic encoder
CN110335233B (en) Highway guardrail plate defect detection system and method based on image processing technology
CN110659550A (en) Traffic sign recognition method, traffic sign recognition device, computer equipment and storage medium
CN112307919B (en) Improved YOLOv 3-based digital information area identification method in document image
Xiang et al. Lightweight fully convolutional network for license plate detection
CN115311507B (en) Building board classification method based on data processing
CN106960196B (en) Industrial video small number recognition method based on template matching and SVM
CN115797813B (en) Water environment pollution detection method based on aerial image
CN111915583A (en) Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene
CN105354547A (en) Pedestrian detection method in combination of texture and color features
CN115033721A (en) Image retrieval method based on big data
CN111027564A (en) Low-illumination imaging license plate recognition method and device based on deep learning integration
CN108509826A (en) A kind of roads recognition method and its system of remote sensing image
CN110349119B (en) Pavement disease detection method and device based on edge detection neural network
CN109829511B (en) Texture classification-based method for detecting cloud layer area in downward-looking infrared image
CN107704864B (en) Salient object detection method based on image object semantic detection

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