Unlock instant, AI-driven research and patent intelligence for your innovation.

Method for detecting adaptive area pooling objects based on SVR model

A technology of object detection and detection method, which is applied in the multimedia field, can solve the problems of slow processing speed, troublesome preprocessing, and inapplicability of non-rigid objects, etc., and achieve the effect of performance improvement and large performance improvement

Active Publication Date: 2017-06-06
JIANGSU COLLEGE OF INFORMATION TECH
View PDF6 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, most of these algorithms use rectangular boxes to shape image blocks, which is not suitable for non-rigid objects
In the prior art, for problems such as illumination changes and occlusion, a moving object detection method using random image selection and adaptive background update is proposed, and the tracking transformation matrix is ​​used to perform adaptive iterative update on the background. However, this method is more suitable for low-speed motion scenes. And the preprocessing is cumbersome, and the processing speed is slow; in the prior art, there is also a 3D image object detection method using an airborne early warning aircraft, which regards the interference in the detection process as a noise factor, and establishes a 3D visual model according to the relevant theory of factorization. A sequence of nonlinear filtering windows is used to filter the 3D visual model; the pyramid pooling method is also used to present object features, but this representation ignores the important geometric information between regions

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method for detecting adaptive area pooling objects based on SVR model
  • Method for detecting adaptive area pooling objects based on SVR model
  • Method for detecting adaptive area pooling objects based on SVR model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0031] Such as figure 1 with figure 2 Shown, a kind of adaptive region pooling object detection method based on SVR model, this detection method comprises the following steps:

[0032] Step 1: Select representative examples;

[0033] One of the researches on multi-instance-based models is to group the training data, and then use the instances in the group as active instances. However, the different appearance of training instances can lead to unsatisfactory clustering results, as instances with low similarity are easily absorbed by the dominant cluster.

[0034] The present invention proposes to find a set of examples of its similar regions. In order to achieve this purpose, the spectral clustering method is adopted first. This method utilizes the pairwise similarity between samples, and then adopts the pyramid pooling method to combine the SIFT histogram The two layers of data are used as appearance features, and the interior between the features is used to form a Laplaci...

Embodiment 2

[0067] Embodiment 2: The detection method can also input the CNN model in the boundary to obtain the features of the training data set blocks, and connect these block features into a feature vector, and then introduce the feature vector value into the SVR model.

[0068] In this embodiment, only the features in the region pooling stage are replaced, and all other steps are kept consistent with those in the experiment in Table 1. Here, instead of pooling the SIFT features of all blocks, the bounding box of each block is used to input the CNN model to obtain features. Then these block features are concatenated into a feature vector, and the comparison results of various aspects are shown in Table 2.

[0069] Table 2: MAP values ​​for each category in the PASCALVOC 2007 test set (section)

[0070] ESVM LDA SPM method Region Pooling the cow 22.7 21.5 40.9 37.5 car 14.1 13.8 32.9 35.4 bike 12.7 12.5 24.7 29.1 the bus 8.9 9.7 19....

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a method for detecting adaptive area pooling objects based on an SVR model. The method for detecting adaptive area pooling objects based on the SVR model is suitable for a partitioned area, an area rough matching mechanism can effectively select and test objects, can automatically discover different samples and areas, and can adapt to area structure changes of objects, an area pooling method is utilized to extract characteristics to analyze an area structure, finally, a non-maximum value suppression method is adopted on categorical data to obtain a detection result, very good effects are achieved on non-rigid targets (such as cows, sheep and the like), compared with other similar methods, the proposed method has obviously improved performance in object detection, the average recall rate reaches 90.8%, CNN characteristics are added, and thus the performance improvement amplitude is larger.

Description

technical field [0001] The invention belongs to the field of multimedia technology, in particular to an SVR model-based self-adaptive region pooling object detection method. Background technique [0002] Affected by blocks, features, and imaging conditions such as vision and noise, the appearance of objects will change greatly, and the visual effect will be good and bad, which brings great challenges to object detection. Since object detection has a wide range of applications in security, battlefield reconnaissance, and agriculture and forestry, it is of great significance and commercial value to design an excellent object detection method. [0003] In order to deal with these changes, the general solution is to use information such as object shape and size, and reduce the impact of noise. In general, similar visual areas exhibit more similar shapes, sizes and structures. By observing these maps, we can further link the region structure and feature extraction. Automatical...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/62
CPCG06F18/2411
Inventor 季云峰
Owner JIANGSU COLLEGE OF INFORMATION TECH