Visual recognition using deep learning (traffic detection using deep learning)

• Main Author: Chen, Sheng
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/154036
• Institution: Nanyang Technological University

In recent years, with the rapid development of autonomous driving, the objective detection technology in driving assistance has become a major research hotspot. Accurate identification of traffic signs from complex traffic scene images is one of the important research directions, which has great significance for maintaining traffic order and reducing traffic accidents. Today, deep learning-based objective detection technology is widely used in various fields. The research of object detection technology has been one of the most basic and challenging research topics in the field of computer vision. In addition, intelligent perception of road environment is an important part of autonomous driving technology, mainly relying on high-resolution cameras, ultrasonic radar, laser radar, GPS locator and other equipment to obtain timely and accurate road signs, potholes, roadblocks, pedestrians and other driving environment information. Generally, the detection accuracy can reach up to 93% (humans can reach about 95%), which cannot meet the expected standard of autonomous driving. Theories and practices show that deep learning algorithms have the ability to perceive complex environments, and the detection accuracy can reach more than 95%. The first part of this project focus on the new acknowledges learning. Related literature and lecture notes have been reviewed. Software/library have been installed and practiced. The second part of this subject is to analyze YOLO sequence, elaborate the research background and significance of target detection, analyze the development of YOLOv1, YOLOv2, YOLOv1YOLOv3, YOLOv4 and YOLOv5 in detail, analyze the current situation of target detection and the significance of YOLO sequence in Object detection. Lastly, using YOLOv5 real-time object detection algorithm to identify and detect traffic signs and vehicles, reduce safety accidents, and assist in driving. Based on the analysis of the principle and characteristics of the yolov5 algorithm, the steps of objective detection using the yolov5 algorithm are presented. The image data is selected by the labelimg software, and the related training and testing sets are sorted out. By enhancing image data pre-processing and optimizing related network parameters, the model convergence is accelerated, and the effect of real-time detection of traffic signs and driving vehicles is achieved. Compared with other traditional detection methods, yolov5 detection results are slightly better. The results show that the traffic sign detection based on yolov5 can meet the needs of real-time and accuracy.