Research On Human Behavior Recognition Technology In Video

As one of the strategic technologies of social informatization,artificial intelligence has received extensive attention from experts and scholars in recent years.As a key research direction of artificial intelligence,human behavior recognition technology has been widely used in many fields such as intelligent monitoring,human-computer interaction,medical assistance and virtual reality.Although the technology has made great research progress,there are still many challenges.For example,in the process of human behavior recognition,the result is easily interfered by external factors such as noise and poor robustness of the algorithm.This paper mainly investigates the research on human behavior recognition algorithms in videos,its specific work is as follows:(1)In the prepossessing of human behavior recognition in video,an improved Canny edge detection algorithm based on hybrid filter and adaptive threshold is proposed to deal with the blur phenomenon of target edges caused by Canny algorithm.With the superiority of bilateral filtering and median filtering in detail retention and noise removal,the mixed filter of bilateral and median filtering is used for filtering processing.In the process of segmenting gradient images,this paper adopts an adaptive threshold calculation method instead of a fixed high and low threshold.The experimental results show that the improved Canny algorithm has good anti-noise characteristics and the number of false edges detected is less.(2)For dealing with incomplete edge information of human targets and the existence of holes in the interior detected by the three-frame difference method,this paper proposes a target detection algorithm based on the Canny edge detection and the three-frame difference method.After drawing the human target in the video sequence by the three-frame difference method,we obtain the three-frame difference image and detect the edge of the three-frame difference image using the improved Canny algorithm.Then by performing the logical "OR" operation with two image and obtaining the public part we get the human body target image.The numerical experiment shows that the algorithm can extract relatively complete human target edge information,the detected target edges are clear and have fewer internal void areas.(3)A human behavior recognition method based on muti-feature fusion is proposed.First,the SURF and HOG features are sparsely coded and sparsely represented by the visual bag-of-words model and then combined with the human contour features in series.Finally,three features are input into the SVM-KNN classifier for classification.The experimental results on Weizmann and KTH databases show that this method can improve the recognition rate of human actions in video and have good robustness.