Research On Human Action Recognition Method In Complex Scene

The problem of human action recognition in simple scene has been solved,but most human action recognition methods only apply to simple scene,applied to the complex scene due to the influence of such factors as clutter background,it is difficult to correctly detect the human action characteristic,a sharp drop in recognition effect,Human action recognition under complex scene became a difficult problem to be solved.In order to detect human action characteristics under the noise interference,this paper designs a multi-scale morphology algorithm in image edge detection for noise resistance,an anti-noise morphological edge detection algorithm is used to detect image edge features,choose wavelet fusion method to improve the effect of fusion,when images are polluted by noise,the method can effectively suppress noise influence on detection results,detect the comparatively accurate edge image.In order to detect human action characteristics in clutter background and camera movement scene,first detect space interesting points using a saliency point detector proposed by Kadir.Then,applying background interesting point suppression to the detected interesting points,reduce the impact of clutter on interest detection.Finally,using 1D Gabor filter in the time domain to obtain space-time interesting points and through the time-domain constraints to reduce the camera movement to examine the effects of point of interest.Based on the BOV model,sparse coding method is used to avoid the disadvantages of k-mean methods,construct a more discriminating code book.The classification effect was further improved by introducing multi-core support vector data description method.Finally,validate our approach by experiments with YouTube,UCF and Hollywood datasets,in order to facilitate comparative classification effect,increase the experiment on the KTH dataset.Then compare the feature detection effect and recognition effect with some commonly methods.Confirmed that our method has a low error detection rate in noise interference,clutter background and camera movement cases,the detection effect is good.The improved human action recognition method proposed in this paper can effectively improve human action recognition rate under complex scene.