Research And Implementation Of Traffic Gesture Dynamic Gesture Recognition Method Based On Neural Network

In recent years,the research fever of driverless technology has been high.In order to comply with the wave of Internet technology development,many scholars have invested in the study of traffic police gesture recognition.At present,the commonly used traffic command methods mainly include fixed traffic signal command and traffic police gesture command.With the growing maturity of driverless technology,driverless cars not only need to have the ability to identify fixed traffic signals,but also respond to and handle complex traffic alert gestures in a timely manner.Based on the existing research,this paper deeply explores the detection and recognition methods of traffic police gestures.The OpenCV library function,gesture detection algorithm and neural network algorithm are used to complete the identification task of traffic police gestures and realize a set of traffic police dynamic gesture recognition.system.The main research work of the thesis is as follows:(1)A variety of traditional gesture detection techniques are used to detect traffic gestures.Different gesture segmentation results are compared.It is found that there are a lot of false edge information and edge contour breaks in the segmented gestures.Based on the above research,a gesture detection technique is proposed.The dynamic region edge point retention algorithm is combined to implement gesture segmentation and gesture contour information extraction.(2)Aiming at the interference of non-rigid problems such as complex background and human motion,gesture detection technology combined with dynamic region edge point integration algorithm is used to realize gesture segmentation and gesture contour information extraction,and effectively eliminate a large number of gesture false edge information and improve extraction.There are problems such as severe fractures in the process.Experiments show that the traditional gesture detection method combined with the dynamic region edge point retention method has a good detection effect,and can also better deal with the recognition problems caused by the lighting environment,the inconsistent gesture speed and the different gesture postures.(3)In view of the problem of low traffic control recognition rate of traffic police,BP neural network and extreme learning machine are used to realize the gesture recognition of traffic police.After comparing the gesture recognition results,it is found that the recognition effect of extreme learning machine is significantly better than BP network..In addition,based on the research of traditional extreme learning machine,combined with particle swarm optimization algorithm,the algorithm of extreme learning machine of neural network is improved.The experiment proves that the improved algorithm has a significant improvement on the gesture recognition effect of the traffic police,and has a high recognition rate and operational efficiency.