Key Technologies of Video-based Animal Behavior Intelligence Analysis System
All along, in the analysis of animal behavior, the behavior of animals is usually observed and analyzed by artificial methods, and then judged. This approach not only consumes time, wastes the researcher's energy, but because of the inevitable subjective factors of the artificial method, it is highly likely that the result of the judgment will be inaccurate or completely opposite to reality. Aiming at this problem, this study takes horse as the experimental object. The problem of animal behavior analysis is summarized as the process of computer acquiring and symbolizing action information by detecting motion data, and then extracting and quantifying action features to realize action behavior classification. This paper mainly discusses and improves the behavior recognition system from three aspects, including the detection of moving targets, the extraction of points of interest and the description of features. The specific contents are as follows: comprehensive and in-depth discussion of animal behavior recognition from interest point extraction to behavior classification The key technology of the stage; in the process of extracting time and space points of interest, the more complex the background of the video, the more likely the detected points of interest fall in the background, causing misjudgment of time and space points of interest and improving the false positive rate of behavior recognition; In order to make the described feature information more abundant, it is proposed to describe the points of interest and their neighborhoods by using multi-feature fusion descriptors. Finally, this paper develops an animal behavior intelligence analysis system that implements the above key technologies.