Research On Traffic Police Gesture Recognition Technology Based On Computer Vision

With the development of technology,cars with assisted driving and unmanned systems frequently appear in people's field of vision.Due to the complexity of urban roads and the serious traffic congestion in China,it is difficult to satisfy traffic control system only relying on fixed indicator signals such as traffic lights and traffic signs.Therefore,a mature autonomous driving system also needs to respond accurately to the traffic police's command gesture.At present,it is the most economical and convenient method to directly identify the collected traffic image of the traffic police command gesture using computer vision technology.Based on the factors such as the accuracy,real-time and anti-interference ability into comprehensive consideration,this paper pays more attention on studying the technology of traffic police gesture recognition depended on Feature-combined technology and Convolutional Neural Networks(CNN)to come over the shortcoming of the current traffic police gesture recognition algorithm from two aspects,which embrace the hand-cranked gesture recognition rate and classification speed in the complex background.The main research contents of this paper are listed as follows:(1)In view of the fact that there is no open source database in the field of traffic police gesture recognition,a small database based on the key actions about Chinese 8 vital group traffic police gesture is established through analysing relative technology,collecting and disposing traffic police gesture image to check out whether the algorithm of identifying traffic police gestures right or not.(2)Facing the problems that the Histogram of Oriented Gradient(HOG)has higher feature dimension and more redundant information,which affects the recognition effect.Principal Component Analysis(PCA)is adopted to reduce the dimensionality of the HOG.As for PCA-HOG features having insufficient ability to describe image information,Local Binary Patterns(LBP)is applied to the algorithm of traffic police gesture recognition,and the improved PCA-HOG feature is combined to obtain the fusion feature operator.And the support vector machine(SVM)is used to construct multi-classifiers.Then,the results show that the recognition algorithm based on feature fusion improves the recognition rate and classification speed of traffic police gesture in complex background.(3)In view of the current application of convolutional neural networks in the field of traffic police gesture recognition,CNN-A and CNN-G are designed for traffic police gesture recognition according to Alex Net and Goog Le Net networks which have some notable achievements in field of image classification.In addition,selecting appropriate network parameters and improving network generalization capabilities to optimize the experimental network.The classification and detection experiments of the features extracted by CNN using Softmax classifier show that the algorithm has a high recognition rate in the traffic police gesture recognition task,and validates the validity and feasibility of the traffic police gesture recognition method based on deep convolutional neural network.Finally,the paper comprehensively compares the above two recognition algorithms.The results show that the improved feature-based fusion traffic gesture recognition algorithm has a good recognition rate in the small-size sample,and the traffic police gesture recognition algorithm based on convolutional neural network has a higher recognition rate in the large-sizes sample.