Research on Behavior Recognition Method Based on Machine Learning and Fisher Vector Coding

Lu, Xing-hua; Yuan, Zi-yue; Lin, Xiao-hong; Qiu, Zi-qi

Aiming at the problem that the existing behavior recognition method can not extract the human body interaction area, resulting in low recognition rate, a behavior recognition method based on machine learning and Fisher vector coding is proposed. Constructing artificial neural network based on machine learning, designing the main steps of backward propagation neural network, making the cost function minimum; using the depth continuity of the image to extract the foreground part of the video motion, multiplying with the corresponding 2D video frame to detect the time domain motion Behavior; Solving the dual quadratic programming problem of Fisher support vector machine, obtaining its optimal solution and completing behavior learning; segmenting the current frame image, solving the normal vector to extract the moving target, and completing the behavior recognition method based on machine learning and Fisher vector coding the study. In order to verify the effectiveness of the design method, a comparative experiment was designed. The experimental results show that the average recognition accuracy of the design method is 7.6% higher than the traditional method.