Graduation Dissertation of Wuhan University of Technology

Research on Behavior Recognition Algorithm Based on Image Processing

(School): Institute of International Education

Specialty amp; Class: Automotive Engineering class 1303

Student name: Jingyu Li

Tutor: Linzhen Nie

Statement of Originality

I solemnly declare that the thesis I submit is a piece of writing I completed independently based on my own research under the guidance of my supervisor. Except for the quotations, this paper does not include any writings published or completed by other individuals or groups. I am fully aware that the legal consequences of this statement are borne by me.

Signature of the author:

Date:

Copyright Letter of Authorization to The Dissertation

The author of this dissertation fully understands provisions stipulated by this school on keeping and using dissertations. The author agrees that the school can reserve and send photocopies and electronic versions of this thesis to relevant departments or institutions managing dissertations. This thesis can be read and borrowed. I hereby authorize excellent provincial institutions appraising bachelor dissertations to incorporate all or any part of this dissertation into relevant databases for retrieval, as well as to preserve and compile this dissertation by photocopying, scanning or reprinting it in a reduced format.

This dissertation is: 1. Confidential until(year) and then this authorization letter will apply. 囗

2. Non-confidential. 囗

(Please tick "√" the corresponding box above)

(Song typeface small, size 4)

Signature of the author:

Date

Signature of the tutor:

Date

Abstract

In recent years, with the rapid development of computer vision, a series of computer vision-related issues has become hot topics of academic research, and behavior recognition technology is one of them. An important part of human behavior analysis technology is based on image video behavior analysis, which has broad application prospects in intelligent video surveillance, human-computer interaction, virtual reality and intelligent driving. At the same time, because the human body movement itself has certain complexity and the surrounding environment has the change, the behavior analysis at this stage also has many problems.

The subject of this paper is the behavior recognition algorithm based on image processing. The article consists of three parts, namely, behavior foreground extraction, behavioral description and human behavior recognition. For the first part, we discuss several foreground extraction algorithms and finally decide to choose the ViBe algorithm. For the second part, the mixed feature of directional gradient histogram and contour moment is used as the method of behavior description. For the third part, the behavior identification problem is set as the classification problem, the advantages and disadvantages of several classification algorithms are studied, and the support vector machine is selected as the behavior recognition classification algorithm.

This paper uses Visual Studio 2013 and Matlab2010b for joint programming. Use the video in the Weizmann database to test the accuracy of the recognition system. Experiments show that this algorithm can identify the human behavior, and the recognition accuracy can reach a satisfactory level.

Key words: behavior recognition, image procession, HOG, SVM, GMM

Cataloge

Ⅰ. Introduction 1

1.1 The Research Background and Significance of the Subject 1

1.2 Research Status and Development Trends at Home and Abroad 3

1.3 The Content of This Paper 5

1.4 The Summary of the Chapter 8

Ⅱ. Research on Target Detection Algorithm 9

2.1 commonly used moving target detection algorithm 9

2.1.1 Optical flow method 9

2.1.2 The method of background subtraction 12

2.1.3 The ViBe Algorithm 15

2.2 Morphological treatment 18

2.3 Video frame image transformation principle and edge extraction algorithm 20

2.3.1 Convolution 20

2.3.2 Point detection 21

2.3.3 Line detection 21

2.3.4 Canny edge detection algorithm 22

2.3.5 Harris angular point detect algorithm 24

2.4 The actual project to realize the foreground detection 26

2.5 The Summary of the chapter 27

Ⅲ. Research of Behavioral Description 28

3.1 Characteristic moments and Hu moments of images 29

3.1.1 Geometric moment 29

3.1.2 Hu moment 30

3.1.3 Contour moment 32

3.2 Histogram of Oriented Gradient 33

3.3 Hog feature simplified method analysis 35

3.3.1 The introduce of the principle of PCA 36

3.3.2 The specific steps of PCA analysis 37

3.4 A Hog feature example analysis and calculation 38

3.5 The Comprehensive Consideration of Feature Selection 39

3.6 Summary of this chapter 40

Ⅳ. Study on Behavior Recognition Algorithms 42

4.1 Template matching algorithm 42