Research And Application Of Traffic Command Gesture Recognition Algorithm In Complex Scene

With the rapid development of Chinese economy,traffic problems such as traffic jams and traffic accidents have become increasingly prominent.Intelligent transportation system is the main means to solve traffic problems.Intelligent vehicle system is an important part of intelligent transportation system.Driverless systems and intelligent assisted driving systems are the main research contents of intelligent vehicles.Traffic command gesture recognition is the key technology of driverless systems and intelligent assisted driving systems,and is also a research hotspot of smart car technology.The traffic environment is easily interfered by factors such as light,vehicle movement,occlusion,and different weather environments,which makes the traffic scene more complicated and difficult to identify traffic command gestures correctly.Therefore,the research on detecting traffic police accurately and identifying traffic command quickly has certain practical significance in complex traffic scenarios.This paper mainly studies traffic command gesture recognition methods in complex scenes based on video information.The main research contents are as follows:1.Under the premise of ensuring the integrity of traffic police information,this paper researches and analyses the common image denoising algorithms and common image enhancement algorithms,and selects bilateral filtering algorithm for denoising,and uses Gamma correction algorithm for image enhancement.2.In order to improve the detection efficiency and detection accuracy of traffic police,a target detection algorithm based on HOG features and SVM classifier was studied.And the target detection algorithm of SSD model and Faster R-CNN model was analyzed and compared,finally selects the Faster R-CNN model as the model of traffic police target detection.Aiming at the situation where there may be multiple traffic policemen,a traffic police localization method combining human aspect ratio and inter-frame difference method is proposed,and the effectiveness of the proposed method is verified through experiments.3.In order to extract the characteristics of the key points of the traffic police skeleton accurately,the skeleton key point extraction algorithm based on the Kinect sensor is analyzed,and the skeleton key point extraction algorithm based on the Open Pose network structure is mainly studied.And for the problem of slow speed of identifying key points of bones by Open Pose,it is optimized by replacing the number and size of convolution kernels.4.In order to recognize the traffic command gestures quickly and accurately,this paper studies the gesture recognition algorithm based on DTW,and focus on the research of LSTM algorithm based on recurrent neural network,the and the gate mechanism of the LSTM network structure is optimized by using jump connections.At the same time,the variational dropout method is added during regularization.Finally,the effectiveness of the optimization method is verified through experiments.