Study On Key Techniques Using In Video Based Animal Behavior Intelligent System

Traditional methods for animal behavior analysis are observing and recording by human. However, such methods not only wasting time and labor resources, but also leading to experimental data deviations even errors due to subjective and inaccurate measurements by human. Aiming at these problems, this dissertation takes white mouse as research object, and studies video-based animal behavior analysis techniques through combination of computer vision, video analysis and pattern recognition technology, among which object detection, object tracking, shape feature extraction and animal posture classification techniques are stressed. In the end, animal behavior analysis software system is developed。The main contributions of this research include:(1)According to the demand of animal observation experiments, a totally schema for animal behavior analysis system is designed, which includes the hardware compositions, a reasonable method for video signal capturing, a technical course for software system. This schema is the fundamental work for building an integrity system.(2)Basing on analysis and comparison of the cost for different background model building methods, time-averaged background image method is proposing in order to build the background image. Experimental results prove that the performance is good when extracting 10 frames in the gap of 3 frames.(3)Analyzing the theory of Camshift algorithm for object tracking in continuously frames, we prompte a Camshift algorithm based on color feature due to the appearance of animal object relatively unique than background colors. Experimental results show that our algorithm can track the animal accurately and quickly, and it is approximately 30ms to 70ms to employ the algorithm in every frame for tracking object. Basing on the Camshift algorithm, a novel object contour extracting method is proposing. Experimental results present that this method is robust and the contours extracting are accurate and clear.(4)Studying on the animal behavior analysis technology, presenting methods for calculating animal moving parameters, we also proposing posture classification method for white mouse based on 1-v-1 SVM technology. In this method, RBF kernel function is selected; 10-fold cross-validation and grid-search method are used for the best parameter selection. After training by 984 samples, the SVM classification precision rate for training set and test set is 98.37% and 86.20% respectively.(5)According to the animal behavior model framework that we designed, a novel animal behavior analysis method model is proposed and is applied in the white mouse daily behavior analysis. The model we proposed accord with natural language and have many advantages: convenient to be employed in program, easy to extend and modify, useful for common applications.(6)An video based animal behavior intelligent analysis system is developed on the basis of Microsoft VC++6.0, Inter OpenCV and Microsoft DirectShow component. This system obtains the animal object moving parameters and posture information quickly and accurately through loading video files or cameras, and provide the animal behavior analysis functions for users, which prove the excellent performance of the key techniques that we study on in the dissertation.