Research On Traffic Sign Real-time Detection Method Based On Improved YOLOv3

With the development and application of machine learning and deep learning in the field of computer vision,more and more successful algorithms can be applied in many industrial fields.For example,the use of successful algorithms can improve the current automotive auxiliary system.The vehicle auxiliary system can make the driver concentrate on the operation of the car on the road,give the judgment of the road information to the system,the judgment results are fed back to the driver,the driver in the corresponding operation,for safe driving.In the road information,the traffic sign is a kind of important information,in order to ensure the safety of driving cars,in the process of high-speed driving fast and accurate detection of the traffic sign information is essential.In the industrial field where target detection has high real-time requirements,YOLOv3 algorithm is the best choice.Especially,YOLOv3 algorithm has been successfully applied in many aspects of intelligent transportation,such as the detection of driving and pedestrians on the road.However,there are some problems in using YOLOv3 directly for target detection test of traffic signs in real environment.For example,the complex background factors around the traffic signs,the small target of the traffic signs and the existence of multiple sizes make it difficult to realize multi-target detection,mistaken and missed detection.In addition,the real-time detection method of traffic signs lacks a solution to meet the market demand in the application field of miniaturized terminals.This thesis will use YOLOv3 to improve the traffic signs,conduct in-depth research on the real-time detection method of traffic signs,and complete the real-time detection system of traffic signs based on edge learning.The main work is as follows:(1)A feature extraction network based on attention mechanism is proposed.This network uses the combination of channel attention and spatial attention in the attention mechanism to embed it into the YOLOv3 feature extraction network,so as to enhance the proportion of the original algorithm’s feature extraction network to the feature extraction of traffic signs in the road environment and reduce the interference of complex background.(2)This thesis proposes a real-time detection method of traffic signs based on improved YOLOv3,TSIG-YOLOv3.This method is based on the yolov3 algorithm of feature extraction network with embedded attention mechanism,and a new detection layer is added to the original YOLOV3 three-layer feature pyramid structure to construct a four-layer new feature pyramid structure,which increases the scale range of YOLOV3 for target detection.Tsign-Yolov3 improves the accuracy of detection of small targets and multi-size traffic signs.(3)The hardware platform of real-time detection system of traffic signs is built,and the real-time detection system of traffic signs based on edge learning is designed and completed.System designed in this paper through the Raspberry Pi 4b camera video image real-time acquisition,through CSI interface will collect video transmission to the Raspberry Pi 4b development board,and the adoption of NCS2 will TSIGN-YOLOv3 method proposed in this paper after the model transformation,deployment on the Raspberry Pi 4b,using edge learning speed of video traffic sign for testing,test results show that the system can meet the requirements of real-time traffic sign detection.