CN107886066A - A kind of pedestrian detection method based on improvement HOG SSLBP - Google Patents
A kind of pedestrian detection method based on improvement HOG SSLBP Download PDFInfo
- Publication number
- CN107886066A CN107886066A CN201711084863.1A CN201711084863A CN107886066A CN 107886066 A CN107886066 A CN 107886066A CN 201711084863 A CN201711084863 A CN 201711084863A CN 107886066 A CN107886066 A CN 107886066A
- Authority
- CN
- China
- Prior art keywords
- scale
- image
- features
- sslbp
- hog
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 40
- 230000004927 fusion Effects 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims description 35
- 238000000034 method Methods 0.000 claims description 30
- 101000822805 Naja atra Cytotoxin A5 Proteins 0.000 claims description 29
- 101000822803 Naja kaouthia Cytotoxin homolog Proteins 0.000 claims description 29
- 101000783567 Naja naja Cytotoxin 1 Proteins 0.000 claims description 29
- 101000822819 Naja naja Cytotoxin-like basic protein Proteins 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 8
- 230000009467 reduction Effects 0.000 claims description 7
- 238000000513 principal component analysis Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000005286 illumination Methods 0.000 abstract description 9
- 238000000605 extraction Methods 0.000 abstract description 6
- 239000000284 extract Substances 0.000 abstract description 3
- 230000008859 change Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 239000013256 coordination polymer Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000003292 diminished effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 239000002904 solvent Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of based on the pedestrian detection method for improving HOG SSLBP, comprise the following steps：Obtain image to be detected sequence；Extract the SSLBP features of above-mentioned image and HOG features and merged, obtain fusion feature；Above-mentioned fusion feature is classified using the HIK SVM classifiers trained, obtains pedestrian detection result.The pedestrian detection method of the present invention has robustness to illumination, noise and rotation, while retaining original image texture information and marginal information, simply and effectively solves scale problem, improve the scale invariability of local binary patterns, obtain scale invariant feature and improve accuracy of detection, and extraction feature is time-consuming shorter, has higher real-time.
Description
Technical Field
The invention relates to the technical field of computer vision and model identification, in particular to a pedestrian detection method based on improved HOG-SSLBP.
Background
Pedestrian detection, as one of object detection techniques, has been widely used in automobile driving assistance, video monitoring systems, and content-based video retrieval. Pedestrian detection can be viewed as a process of feature extraction combined with classifier design in order to automatically analyze and detect the presence of pedestrians from an unknown video or image. By adopting the intelligent image information analysis technology of pedestrian detection, the uncertainty problem in image understanding and pattern recognition related to target detection can be effectively processed.
At present, more classical pedestrian detection methods include a pedestrian detection method combining an HOG descriptor and an SVM classifier, a method for detecting a pedestrian with a serious adhesion phenomenon by using a variable part model (DPM), and a pedestrian detection method based on an HOG-LBP feature. The use of HOG and its improved method has met with great success. The HOG and the LBP are fused to be used as characteristics for pedestrian detection, and the defect of edge direction information loss caused by singly using the LBP characteristics for image description is overcome. And the HOF and the local self-similarity feature are used for image description, and the HIKSVM classifier is used for pedestrian detection, so that the accuracy and the real-time performance of the algorithm are improved.
When the Local Binary Pattern (LBP) is used as an operator for describing local texture features of an image and used for extracting the texture features, the LBP has the remarkable advantages of rotation invariance, gray scale invariance and the like, and the extracted features are the local texture features of the image. But has the defects of higher feature spectrum dimension, poor robustness to illumination, rotation and scale and the like.
The HOG descriptor generation process is long, so that the speed is low, the real-time performance is poor, and the problem of shielding is difficult to process. Due to the nature of the gradient, the descriptor is quite sensitive to noise. The LBP and the improved operator thereof can not solve the problem of scale change, and have the defects of higher characteristic spectrum dimension, poor noise robustness and the like.
Disclosure of Invention
In view of the above analysis, the present invention aims to provide a pedestrian detection method based on improved HOG-SSLBP, so as to solve the problem that the existing pedestrian detection method cannot solve the scale change problem or has poor real-time performance.
The purpose of the invention is mainly realized by the following technical scheme:
a pedestrian detection method based on improved HOG-SSLBP specifically comprises the following steps:
acquiring an image sequence to be detected;
extracting and fusing the SSLBP characteristic and the HOG characteristic of the image sequence to obtain a fusion characteristic;
and classifying the fusion characteristics by using a trained HIK SVM classifier to obtain a pedestrian detection result.
The invention has the following beneficial effects:
the invention adopts a feature extraction method of fusing Scale Selection Local Binary Pattern (SSLBP) features of an original image with HOG features to extract image features. The method has robustness to illumination, noise and rotation, and retains original image texture information and edge information; the scale problem is simply and effectively solved, the scale invariance of the local binary pattern is improved, the scale invariance characteristic is obtained, the detection precision is improved, the time consumption for extracting the characteristic is short, and the real-time performance is high.
On the basis of the scheme, the invention is further improved as follows:
further, the extracting the SSLBP features of the image sequence specifically includes the following steps:
extracting scale sensitive features by using CLBP;
and processing the obtained scale sensitive characteristics by applying a scale selection scheme to obtain the SSLBP characteristics.
The beneficial effect of adopting the further scheme is that:
and the CLBP is used for extracting scale sensitive features, so that the influence of rotation is eliminated.
Further, the scale selection scheme specifically comprises the following steps:
obtaining a scale space of an image using a gaussian filter;
constructing a local mode histogram of each image only containing the dominant mode in the scale space;
for each dominant mode, selecting a maximum frequency of occurrence between different scales;
and constructing and obtaining a new feature histogram by using the selected maximum occurrence frequency, wherein the new feature histogram is used as the scale-invariant feature of the image.
Further, the dominant mode is determined by:
obtaining a scale space of a training sample set through a two-dimensional Gaussian filter;
constructing a local mode histogram for the image in each scale space;
for each mode, selecting a maximum frequency of occurrence between different scales;
constructing and obtaining a new feature histogram by using the selected maximum occurrence frequency;
and performing descending classification on the obtained new feature histogram, and selecting a mode with high average frequency as a dominant mode.
Further, the training process of the HIK SVM classifier specifically comprises the following steps:
inputting a training sample set;
extracting and fusing the SSLBP characteristic and the HOG characteristic of the training sample to obtain a fusion characteristic;
and training the HIK SVM classifier by using the fusion characteristics.
Further, the extracting the SSLBP feature of the training sample specifically includes the following steps:
extracting scale sensitive features of the training sample by using CLBP;
and processing the obtained scale sensitive characteristics by using a scale selection scheme to obtain the SSLBP characteristics of the training sample.
Further, the scale-sensitive feature is extracted by using the CLBP, and specifically, a two-dimensional joint histogram "CLBP _ S/C" or "CLBP _ M/C" is established as the scale-sensitive feature.
Further, the extracting the HOG features specifically includes the following steps:
carrying out normalization processing on an image to be detected;
calculating the gradient of the image after the normalization processing by using first order differential;
performing direction weight projection based on the gradient amplitude obtained by calculation to obtain a feature vector;
and carrying out normalization processing on the feature vectors to obtain the HOG features.
Further, the fusion characteristics are obtained, and specifically, the extracted SSLBP characteristics and HOG characteristics of the image to be detected are fused in series.
The beneficial effect of adopting the further scheme is that:
the two image features are fused, so that any texture information and edge information are not lost, and robustness is provided for illumination, rotation, noise and scale.
Further, for the obtained fusion features, the principal component analysis method is used for dimensionality reduction.
The beneficial effect of adopting the further scheme is that:
the characteristic dimension can be greatly reduced, the operation efficiency is improved, and the real-time performance of pedestrian detection is improved.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
Fig. 1 shows a flow chart of a pedestrian detection method based on improved HOG-SSLBP.
Fig. 2 shows a schematic diagram of the HOG feature generation process.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
The invention discloses a pedestrian detection method based on improved HOG-SSLBP. The method comprises the following steps:
and step S1, acquiring the image sequence to be detected.
And acquiring an image sequence to be detected, and preprocessing the image to improve the quality of the image to be detected.
Step S2, extracting image features using modified SSLBP
Specifically, a CLBP is used for extracting scale-sensitive features; a scale selection scheme is applied to obtain SSLBP features.
Step S201, using CLBP to extract scale sensitive features.
CLBP is a gray-scale texture operator that characterizes the local spatial structure of image texture. Giving the central pixel gcAnd P circularly uniformly spaced adjacent gpThe radius is R, P is 0, 1. G can be simply calculatedcAnd gpDifference between because of dp＝gp-gc。dpCan be further broken down into two components:
wherein,is a function of the sign, mpIs dpThe amplitude of (c).
Thus, three operators are defined as:
CLBP_CP，R＝t(gc，cI)
wherein t is a threshold functioncmIs m for an imagepAverage value of cIIs the average gray level of the image. CLBP _ S can be obtainedP，RAnd CLBP _ MP，RAll can have 2pA different value. Since CLBP _ C contains only one byte, it can have two values that are rotation invariant.
To eliminate the effect of rotation, i.e. to assign a unique identifier for each rotation-invariant pattern, the rotation-invariant quantities CLBP _ S and CLBP _ M are defined as:
wherein, ROR (x, i) is circularly right-shifted i times on P byte number x. Much smaller than CLBP _ SP，R(CLBP_MP，R)。
Since CLBP _ S, CLBP _ M and CLBP _ C generate binary strings and contain supplementary information, scale-sensitive features are extracted by creating a two-dimensional joint histogram "CLBP _ S/C" or "CLBP _ M/C" to obtain more information and avoid huge dimension problems.
Step S202, a scale selection scheme is applied to obtain SSLBP, and scale invariant features are extracted.
When an image is zoomed in or out, there is still a dominant mode, but it will occupy a larger or smaller area of the image. I.e. the percentage of such a pattern in the image is not changed, as is the characteristic scale of the pattern. Thus, by finding the characteristic scale of a pattern and extracting its percentage over that scale, scale invariance can be achieved. In the embodiment of the invention, a novel and simple scale invariant feature extraction method is designed for LBP by determining the dominant mode of a scale space, and the method comprises the following steps: finding out a dominant mode by analyzing a scale space of a training set and applying a training stage; and extracting scale-invariant features of the training and testing set based on the dominant mode.
The dominant mode is found by applying a training phase by analyzing the scale space of the training set. The method specifically comprises the following steps: giving a training sample set, obtaining a scale space through a two-dimensional Gaussian filter, and constructing multiple scales; constructing a local mode histogram for the image in each scale space; for each pattern, the maximum frequency of occurrence of the pattern between different scales is preserved to construct a new histogram that serves as a scale-invariant feature for a given training sample; the dominant mode is determined by selecting a high average frequency in the entire training set.
The scale invariant feature extraction method extracts frequency information from learning and fixed patterns, thus preserving pattern types that contain important information. For two images I1And I2The characteristics will be obtained according to the above method:andwherein P isAAnd PBIs twoThe dominant mode of the type.
In particular, for the above obtainedDetermining the dominant mode at different scales, specifically comprising the following steps:
step S20201, initialize a pattern histogram for the training sample setWhere HS is the size of the rotation invariant histogram. The training sample set is as follows: t ═ fi1, 2., N }, where fiIs a training image and N is the training set size.
Step S20202, for the image f in the training sample setiAnd obtaining a scale space through a two-dimensional Gaussian filter.
In particular, gσFor two-dimensional Gaussian filter with standard deviation sigma for constructing scale space, based on gσFor image fiObtaining a scale space sl。
Where L is the size of the scale space for a given image,
step S20203, for the scale space S obtained abovelEstablishing a mode histogram based on CLBP _ S/C
Step S20204, selecting the maximum frequency of occurrence of the mode in the mode histogram, and constructing an image fiA new histogram of (2):
in the formula, k is more than or equal to 0 and less than or equal to HS-1, and HS is the size of the rotation invariant histogram.
Step S20205, adding the value of the new histogram to
In the formula, k is more than or equal to 0 and less than or equal to HS-1, and HS is the size of the rotation invariant histogram.
Step S20206, repeating steps S20202 to S20205 until all images in the training sample are processed.
Step S20207, for the above obtainedSorting in descending order to obtain K patterns with the first K maximum frequenciesK is the number of dominant patterns learned for CLBP _ S/C.
For theThe specific steps for determining the dominant mode at different scales are similar to the above steps, and are not described herein again.
And step S203, after the dominant mode is determined, extracting the scale invariant feature of the test image sequence. First, given a sample, a scale space is obtained by a two-dimensional gaussian filter. Then, for each image in the scale space, a local pattern histogram is constructed that contains only the pre-learned dominant patterns. And finally, for each dominant mode, selecting the maximum occurrence frequency of different scales to construct a new feature histogram, and extracting the scale to select local binary pattern features.
In particular, by the aboveExtracting scale selection local binary pattern features, specifically comprising the following steps:
step S20301, obtaining the scale space of given sample by two-dimensional Gaussian filter based on gσObtaining a scale space s for an image IlWherein g isσIs a two-dimensional gaussian filter with standard deviation sigma for constructing a scale space.
Step S20302, for the above scale space SlEstablishing a mode histogram based on CLBP _ S/CSpecifically, for each pixel, a mode tag is first computed based on CLBP _ S/C if its mode tag does not belong toThen the pixel will not be rightIt is helpful.
Step S20303, establishing the scale selection feature of the image I:
in the formula, K is more than or equal to 1 and less than or equal to K.
By passingThe specific method and process for extracting the scale selection features are similar to the steps described above, and are not repeated here。
Cascading the two obtained features, and extracting the scale-invariant features of the image:
in the traditional method, most of the invariant features of local or global scale are extracted. From an implementation point of view, the invention is proposed to belong to an intermediate path. The invention firstly extracts local scale change characteristics and then applies global transformation to realize scale invariance. On the other hand, from the perspective of scale space, the present invention proposes a new method.
Step S3, extracting features using histogram of oriented gradients
The HOG (Histogram of Oriented gradient) is an image descriptor for solving the human body target detection, and in this implementation, features of a Histogram of gradient directions (HOG) are used to express a human body, and appearance information and motion information of the human body are extracted to form a rich feature set. As shown in fig. 2, the method specifically includes the following steps:
step S301, normalization processing is performed on the input image. The normalization of the image aims to improve the robustness of the detector to illumination, the normalization of the color space is to normalize the color information of the whole image so as to reduce the influence of different illumination and background, and the Gamma and the normalization of the color space of the image are introduced as the preprocessing means of feature extraction in order to improve the robustness of detection.
Step S302, calculating image gradient by using first order differential. Image edges are caused by abrupt changes in local features of the image, including gray scale, color, and texture. The change between adjacent pixel points in an image is less, the change of the area is flat, the gradient amplitude is smaller, otherwise, the gradient amplitude is larger. The gradient corresponds to the first derivative of the image. The gradient of any pixel point (x, y) in the image f (x, y) is:
in the formula, GxIs a gradient in the x-direction, GyIs the gradient along the y direction, x, y are the horizontal and vertical coordinate value of the pixel point, the amplitude and the direction angle of the gradient are respectively expressed as:
in the formula, GxIs a gradient in the x-direction, GyIs the gradient along the y direction, and x and y are the horizontal and vertical coordinate values of the pixel points.
Because the calculation amount of the modulus is large, the following formula is generally used for approximate solution:
in the formula, GxIs a gradient in the x-direction, GyIs the gradient in the y-direction.
In this embodiment, the gradient and direction of the image are calculated by using the template [ -1, 0, 1], and the gradients in the horizontal and vertical directions are calculated by using the gradient template as follows:
in the formula, Gx，GyAnd the gradients of the pixel points (x, y) in the horizontal direction and the vertical direction are respectively expressed, H (x, y) expresses the gray value of the pixel points (x, y), and the amplitude and direction calculation formula of the gradients is as follows:
for range limitation of the gradient direction, an unsigned range is used, so the gradient direction can be expressed as:
and step S303, projecting direction weight based on the gradient amplitude of the pixel.
Dividing the whole target window into cell units (cells) with different sizes, and calculating the gradient information of each cell respectively, wherein the cell units comprise gradient sizes and gradient directions. The gradient directions of the pixels are evenly divided into 9 bins within the interval of 0-180 degrees, when the gradient directions exceed 9 bins, the detection performance is not obviously improved, but the detection operand is increased, the pixels in each cell perform weighted voting for the histogram of the gradient direction in which the pixels are located, and the weighted weight is the gradient amplitude of the pixels.
And step S304, carrying out normalization processing on the feature vectors.
Specifically, the local cell units are normalized, and the absolute value of the gradient amplitude is reduced, so that the absolute value is easily influenced by the contrast between the foreground and the background and the local illumination, and a more accurate detection effect is obtained.
Combining a plurality of cell units (cells) into a larger block (block), wherein the whole image can be regarded as a window to be detected, the larger block can be regarded as a sliding window, and the larger block can slide from left to right and from top to bottom in sequence to obtain blocks with repeated cell units and gradient information of the same cell units (cells) in different blocks (blocks).
The block information is normalized, and the final detection effect is affected by the sizes of different cell units and different blocks. In the present embodiment, the block size is 3 × 3 cells, and the cell size is 6 × 6 pixels, where the detection effect is the best. If the size of the chunk is too large, the effect of the normalization is diminished resulting in an increased error rate, and if the size of the chunk is too small, useful information is filtered out instead.
This example usesAnd (epsilon is a small constant and avoids the denominator being 0), and the HOG feature vectors in the block blocks are normalized, so that the feature vector space has robustness to illumination, shadow and edge change.
A high-dimensional vector consisting of β × ζ × η data is obtained, that is, a description vector of the HOG for the image is obtained, where β represents the number of directional units (bin) in each cell, and ζ and η represent the number of blocks and the number of cells in one block, respectively.
And step S4, fusing the obtained SSLBP characteristic and HOG characteristic.
Let SSLBP feature vectors of the obtained image be expressed as: x1＝[X11，X12，...，X1n]The HOG feature vector of the image is represented as: x2＝[X21，X22，...，X2m]Then the feature vector after serial fusion is: x ═ X11，X12，...，X1n，X21，X22，...，X2m]。
After obtaining the SSLBP feature of the original image and the HOG feature of the original image are fused in a serial fusion mode, the final image fusion feature vector can be obtained.
And step S5, performing dimensionality reduction on the obtained fusion feature vector by using PCA.
After the final image fusion feature vector is obtained, the Principal Component Analysis (PCA) method is used for reducing the dimension so as to reduce the calculated amount and improve the real-time property of detection. In particular, the amount of the solvent to be used,
given m d-dimensional spatial samples x1，x2，...，xmAnd calculating a covariance matrix:
wherein m is the number of spatial samples
Decomposing the covariance matrix S to obtain a matrix U composed of eigenvectors corresponding to the first k largest eigenvaluesk＝[u1，u2，...，uk]And obtaining the characteristics after dimensionality reduction:
and step S6, classifying the features after dimensionality reduction by using a trained HIKSVM classifier to obtain a pedestrian detection result.
The training process of the HIKSVM classifier comprises the following steps: inputting a training sample set; extracting and fusing the SSLBP characteristic and the HOG characteristic of the training sample to obtain a fusion characteristic; and training the HIK SVM classifier by using the fusion characteristics. The extracting and fusing the SSLBP feature and the HOG feature of the training sample is similar to the extracting and fusing the SSLBP feature and the HOG feature of the test image in the above-mentioned content, and specific content is not described again here. And using the trained HIKSVM classifier for the feature classification obtained after the dimensionality reduction to obtain a pedestrian detection result of the image.
In summary, the embodiment of the present invention provides a pedestrian detection method based on improved HOG-SSLBP, which uses an image descriptor fused with HOG and SSLBP, so as to enhance scale invariance and rotation invariance, have good robustness to illumination, rotation, noise and scale, and improve detection accuracy; meanwhile, the PCA is used for dimensionality reduction, so that the characteristic dimensionality can be greatly reduced while original image texture information and edge information are kept, the operation efficiency is improved, and the real-time performance of pedestrian detection is improved.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. A pedestrian detection method based on improved HOG-SSLBP is characterized by comprising the following steps:
acquiring an image sequence to be detected;
extracting and fusing the SSLBP characteristic and the HOG characteristic of the image sequence to obtain a fusion characteristic;
and classifying the fusion characteristics by using a trained HIK SVM classifier to obtain a pedestrian detection result.
2. The method according to claim 1, wherein said extracting SSLBP features of the sequence of images comprises in particular the steps of:
extracting scale sensitive features by using CLBP;
and processing the obtained scale sensitive characteristics by applying a scale selection scheme to obtain the SSLBP characteristics.
3. The method according to claim 2, wherein the scale selection scheme comprises in particular the steps of:
obtaining a scale space of an image using a gaussian filter;
constructing a local mode histogram of each image only containing the dominant mode in the scale space;
for each dominant mode, selecting a maximum frequency of occurrence between different scales;
and constructing and obtaining a new feature histogram by using the selected maximum occurrence frequency, wherein the new feature histogram is used as the scale-invariant feature of the image.
4. The method of claim 3, wherein the dominant mode is determined by:
obtaining a scale space of a training sample set through a two-dimensional Gaussian filter;
constructing a local mode histogram for the image in each scale space;
for each mode, selecting a maximum frequency of occurrence between different scales;
constructing and obtaining a new feature histogram by using the selected maximum occurrence frequency;
and performing descending classification on the obtained new feature histogram, and selecting a mode with high average frequency as a dominant mode.
5. The method according to any one of claims 1 to 4, wherein the training process of the HIK SVM classifier specifically comprises the following steps:
inputting a training sample set;
extracting and fusing the SSLBP characteristic and the HOG characteristic of the training sample to obtain the fusion characteristic of the sample;
and training the HIK SVM classifier by using the fusion characteristics.
6. The method of claim 5, wherein the extracting the SSLBP features of the training samples comprises the steps of:
extracting scale sensitive features of the training sample by using CLBP;
and processing the obtained scale sensitive characteristics by using a scale selection scheme to obtain the SSLBP characteristics of the training sample.
7. The method according to claim 4, characterized in that the use of CLBP to extract scale-sensitive features, in particular to build a two-dimensional joint histogram "CLBP _ S/C" or "CLBP _ M/C" as scale-sensitive features.
8. The method according to claim 7, wherein said extracting the HOG features comprises in particular the steps of:
carrying out normalization processing on an image to be detected;
calculating the gradient of the image after the normalization processing by using first order differential;
performing direction weight projection based on the gradient amplitude obtained by calculation to obtain a feature vector;
and carrying out normalization processing on the feature vectors to obtain the HOG features.
9. The method according to claim 8, wherein the fusion features are obtained, in particular SSLBP features and HOG features of the extracted image to be detected are fused in series.
10. The method of claim 9, wherein for the obtained fusion features, principal component analysis is used for dimensionality reduction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711084863.1A CN107886066A (en) | 2017-11-07 | 2017-11-07 | A kind of pedestrian detection method based on improvement HOG SSLBP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711084863.1A CN107886066A (en) | 2017-11-07 | 2017-11-07 | A kind of pedestrian detection method based on improvement HOG SSLBP |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107886066A true CN107886066A (en) | 2018-04-06 |
Family
ID=61778994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711084863.1A Pending CN107886066A (en) | 2017-11-07 | 2017-11-07 | A kind of pedestrian detection method based on improvement HOG SSLBP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107886066A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108549868A (en) * | 2018-04-12 | 2018-09-18 | 中国矿业大学 | A kind of pedestrian detection method |
| CN108734139A (en) * | 2018-05-24 | 2018-11-02 | 辽宁工程技术大学 | Feature based merges and the newer correlation filtering tracking of SVD adaptive models |
| CN109919960A (en) * | 2019-02-22 | 2019-06-21 | 西安工程大学 | A kind of image continuous boundary detection method based on Multiscale Gabor Filters device |
| CN110163161A (en) * | 2019-05-24 | 2019-08-23 | 西安电子科技大学 | Multiple features fusion pedestrian detection method based on Scale invariant |
| CN113627520A (en) * | 2021-08-09 | 2021-11-09 | 中南大学 | Scale selection and noise robustness improved local binary pattern texture description method |
| CN114387619A (en) * | 2021-12-31 | 2022-04-22 | 歌尔科技有限公司 | Pedestrian detection method, device, electronic equipment and computer-readable storage medium |
| CN118196479A (en) * | 2024-02-26 | 2024-06-14 | 国网江苏省电力有限公司南京供电分公司 | GIS equipment partial discharge type determining method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663421A (en) * | 2012-03-23 | 2012-09-12 | 天津天地伟业物联网技术有限公司 | Vehicle detection method |
| CN103617426A (en) * | 2013-12-04 | 2014-03-05 | 东北大学 | Pedestrian target detection method under interference by natural environment and shelter |
| CN105447503A (en) * | 2015-11-05 | 2016-03-30 | 长春工业大学 | Sparse-representation-LBP-and-HOG-integration-based pedestrian detection method |
| US20160180158A1 (en) * | 2014-12-18 | 2016-06-23 | Magna Electronics Inc. | Vehicle vision system with pedestrian detection |
-
2017
- 2017-11-07 CN CN201711084863.1A patent/CN107886066A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663421A (en) * | 2012-03-23 | 2012-09-12 | 天津天地伟业物联网技术有限公司 | Vehicle detection method |
| CN103617426A (en) * | 2013-12-04 | 2014-03-05 | 东北大学 | Pedestrian target detection method under interference by natural environment and shelter |
| US20160180158A1 (en) * | 2014-12-18 | 2016-06-23 | Magna Electronics Inc. | Vehicle vision system with pedestrian detection |
| CN105447503A (en) * | 2015-11-05 | 2016-03-30 | 长春工业大学 | Sparse-representation-LBP-and-HOG-integration-based pedestrian detection method |
Non-Patent Citations (5)
| Title |
|---|
| SONGZHI SU,SHUYUAN CHEN.: "Analysis of Feature Fusion Based on HIK SVM and Its Application for Pedestrian Detection", 《ABSTRACT AND APPLIED ANALYSIS》 * |
| ZHENHUA GUO,XINGZHENG WANG,JIE ZHOU,JANE YOU.: "Robust Texture Image Representation by Scale Selective Local Binary Patterns", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 张笑钦 等: "《复杂场景下目标跟踪的理论与方法》", 31 May 2017 * |
| 章霄 等: "《数字图像处理技术》", 31 July 2005 * |
| 贾永红 等: "《数字图像处理技巧》", 31 January 2017 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108549868A (en) * | 2018-04-12 | 2018-09-18 | 中国矿业大学 | A kind of pedestrian detection method |
| CN108734139A (en) * | 2018-05-24 | 2018-11-02 | 辽宁工程技术大学 | Feature based merges and the newer correlation filtering tracking of SVD adaptive models |
| CN108734139B (en) * | 2018-05-24 | 2021-12-14 | 辽宁工程技术大学 | Correlation filtering tracking method based on feature fusion and SVD self-adaptive model updating |
| CN109919960A (en) * | 2019-02-22 | 2019-06-21 | 西安工程大学 | A kind of image continuous boundary detection method based on Multiscale Gabor Filters device |
| CN110163161A (en) * | 2019-05-24 | 2019-08-23 | 西安电子科技大学 | Multiple features fusion pedestrian detection method based on Scale invariant |
| CN113627520A (en) * | 2021-08-09 | 2021-11-09 | 中南大学 | Scale selection and noise robustness improved local binary pattern texture description method |
| CN113627520B (en) * | 2021-08-09 | 2023-09-05 | 中南大学 | Texture description method for improved local binary mode of scale selection and noise robustness |
| CN114387619A (en) * | 2021-12-31 | 2022-04-22 | 歌尔科技有限公司 | Pedestrian detection method, device, electronic equipment and computer-readable storage medium |
| CN118196479A (en) * | 2024-02-26 | 2024-06-14 | 国网江苏省电力有限公司南京供电分公司 | GIS equipment partial discharge type determining method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tong et al. | Salient object detection via bootstrap learning | |
| CN107886066A (en) | A kind of pedestrian detection method based on improvement HOG SSLBP | |
| Al-Shemarry et al. | Ensemble of adaboost cascades of 3L-LBPs classifiers for license plates detection with low quality images | |
| Ranzato et al. | Automatic recognition of biological particles in microscopic images | |
| Vishwakarma et al. | A unified model for human activity recognition using spatial distribution of gradients and difference of Gaussian kernel | |
| Briggman et al. | Maximin affinity learning of image segmentation | |
| Kim et al. | Color–texture segmentation using unsupervised graph cuts | |
| Jia et al. | Visual tracking via coarse and fine structural local sparse appearance models | |
| CN105488809A (en) | Indoor scene meaning segmentation method based on RGBD descriptor | |
| Soni et al. | Text detection and localization in natural scene images based on text awareness score | |
| US10810433B2 (en) | Method and system for tracking objects | |
| CN112784722B (en) | Behavior identification method based on YOLOv3 and bag-of-words model | |
| CN107886067A (en) | A kind of pedestrian detection method of the multiple features fusion based on HIKSVM graders | |
| CN110738672A (en) | image segmentation method based on hierarchical high-order conditional random field | |
| Zhou et al. | Fast circle detection using spatial decomposition of Hough transform | |
| Araar et al. | Traffic sign recognition using a synthetic data training approach | |
| CN108549868A (en) | A kind of pedestrian detection method | |
| Perrotton et al. | Automatic object detection on aerial images using local descriptors and image synthesis | |
| Wang et al. | Cascading classifier with discriminative multi-features for a specific 3D object real-time detection | |
| Neycharan et al. | Edge color transform: a new operator for natural scene text localization | |
| Lu et al. | Superpixel level object recognition under local learning framework | |
| Zhang et al. | Research on vehicle object detection method based on convolutional neural network | |
| Fan et al. | Robust visual tracking via bag of superpixels | |
| Kosala et al. | MSER-Vertical Sobel for Vehicle Logo Detection | |
| CN113128487A (en) | Dual-gradient-based weak supervision target positioning method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180406 |