Real-time Intelligent Gesture Recognition In Complex Backgrounds

The advancement and improvement of science and technology make human-computer interaction appear to be extraordinarily important and important.The flexibility and convenience of gestures make it gradually become the main bridge for human-computer interaction.However,gesture recognition in natural and complex environments is interfered by many factors,making gesture detection and recognition has always been a challenging interdisciplinary research problem.This paper builds a set of gesture detection and recognition interactive control system in indoor natural scenes,and completes the detection and recognition tasks for four static gestures in natural scenes.This paper conducts research in the following two directions:(1)The gesture pattern is quickly segmented by an Adaboost detector combined with Haar features.Due to the diversity of gestures and the complexity of the image background,this paper chooses to combine the TPLBP and HOG gradient histogram features as new features.Identification of smart gestures.Because the HOG operator is robust to dark illumination changes in complex scenes,the TPLBP operator has better extraction performance for gesture texture features in complex scenes.Combining the HOG feature with the TPLBP feature can effectively express the texture features of the gesture and increase the resistance to light pollution under complex conditions.Finally,the merged features are input into the artificial neural network for verification and classification of gestures.(2)Applying the YOLO v3 convolutional neural network,which is gradually hot and skilled in recent years,to detect four gestures in a natural complex environment to realize an end-to-end detection network.However,at present,there are still some problems in YOLO v3: one stage detection target position is not accurate enough,IOU optimization method is not reasonable enough,and deep learning detection rate is low.According to the existing problems,this paper proposes to solve the problems in practical applications by using Focal loss context classification optimization,GIOU regression loss optimization and Tiny-Model frame rate improvement.Finally,based on the Windows platform,an intelligent gesture control system based on human-computer interaction is built.Through the control system,we can complete the human-computer interaction experience of four gestures,and realize “all in the palm”.