Research On Recognition Method Of Traffic Signs Based On Unmanned Vehicles In Complex Environment

Traffic sign recognition technology is one of the important components of unmanned driving.However,the existing unmanned traffic sign recognition systems are often used in closed experimental sections or urban sections,and it is difficult to apply them to remote rural roads and relatively Driving freely between two places far away is because there are more complicated environmental factors than urban roads and more uncertain and complex environmental factors in remote rural areas and between two places far away,so that the traffic sign recognition system The obtained image features are more obvious than those in urban areas,so that the accuracy and real-time performance of the traffic signs recognized by the system cannot meet the higher requirements of unmanned driving.In response to this problem,this thesis uses a convolutional neural network algorithm to realize the recognition of traffic signs in a complex environment.The main contents of the work are as follows:Aiming at the problem of unclear traffic sign images in complex environments,which makes feature extraction difficult,this thesis improves the Inception module in the Google Net network to create the MY＿module module.The MY＿module module uses three branches to use different convolution kernels to perform image feature extraction in parallel.The convolution kernel in the first branch is used to extract the main features of the image;The second and third branches are used to extract the horizontal and vertical features of the image as auxiliary branches;the features extracted from each branch are fused to avoid the loss of an important feature,so that the module can effectively extract the image features of traffic signs in complex environments,and use Leak Relu activation function and BN optimization algorithm to further improve the performance of the module to process data.This module is used to create the New Net network structure and verify it in the GTSRB data set.Experiments have verified that the MY＿module module has a stronger ability to extract traffic sign features in complex environments than the traditional convolution method.Aiming at the problem of detecting traffic signs in real-time complex environment road scenes,the YOLOV3 algorithm is used to detect traffic signs in the natural environment,and three improvements are made to the YOLOV3 model: the backbone network in YOLOV3 is improved,the residual module are replaced in the trunk with the created MY＿module module,the reason is that MY＿module can improve the ability to extract image features,but it has a certain degree of complexity,in order not to affect the real-time performance,the depth of the backbone network is appropriately reduced,adding a detection scale Y4 can improve the detection effect of small targets;K-means++ algorithm is used to replace K-means algorithm,K-means++ can recluster suitable prior boxes;the loss function is improved to increase the improved YOLOV3’s detection rate for small targets,and reduce the impact of large targets on small targets in the same screen.The improved YOLOV3 uses the amplified GTSDB data for experimental verification.The simulation results show that the improved YOLOV3 model is better than the original YOLOV3 model in terms of accuracy and real-time performance in the recognition of traffic signs in complex environments.