Research On Target Detection Algorithm For Traffic Signs

Traffic sign detection method is the core technology of unmanned driving and advanced assisted driving,and it is also the research hotspot of academic circles in recent years.When driving,the traffic sign detection system carries out real-time detection and sends back road condition information,and the driver makes accurate judgment on the current section of traffic condition according to the information,so as to prevent the occurrence of traffic accidents.Traditional target detection algorithms extract traffic sign features through manual labeling,which has low accuracy and efficiency,and it is difficult to guarantee real-time detection in actual road scenes.When collecting traffic sign data,it is inevitably affected by illumination,weather and other factors,or the traffic sign is blocked by trees,which affects the quality of the collected data.In addition,traffic signs generally account for a small proportion in the whole image,so the feature information that can be extracted during detection is less.All these problems increase the difficulty of the traffic sign detection.This paper focuses on the above problems and does the following work:First of all,in view of the low accuracy and speed of traffic sign detection,this paper builds a traffic sign detection model R-YOLOV4 based on YOLOV4 algorithm and combined with the data characteristics of traffic sign data set.The model proposed an improved K-means++clustering algorithm to achieve accurate labeling of the prior box,carried out concat operation and convolution operation on the shallow and deep feature information of the trunk feature extraction network CSPDarkNet53 to achieve feature fusion and output,and completed multi-scale predictionExperimental results show that the MAP value of the improved R-YOLOV4 model is 80.6%,which is 10%higher than that of the original YOLOV4 model.Therefore,the improved R-YOLOV4 algorithm structure can effectively improve the detection performance of traffic signs.Secondly,in order to reduce the computational complexity of the model and accelerate the detection speed of the model,meet the requirements of practical applications and facilitate the transplantation of embedded devices,this paper lightweight R-Yolov4 model and put forward a ligtweight traffic sign detection model-MR-Yolov4 model:The backbone feature extraction network CSPDarkNet53 of R-YOLOV4 model is replaced by MobileNet3,and the lightweight backbone feature extraction network is constructed.The deep separable convolution is used to replace the ordinary convolution in the enhanced feature extraction network,which reduces the computational burden of the whole model.The results show that,compared with the weight of R-YOLOV4 model,the model size of MR-YOLOV4 model is reduced by 5 times to only 2.7MB,which is suitable for deployment on mobile devices.Finally,taking MR-YOLOV4 network as the detection model,a traffic sign detection software system is designed and implemented.Through data acquisition,data preprocessing and traffic sign detection based on MR-YOLOV4 model,the construction of traffic sign detection system is completed.The results show that,compared with the existing traffic sign detection technology,the improved method in this paper has significantly improved the speed and accuracy of traffic sign detection,and has better reliability and robustness.