Optimization of Real-Time Detection and Tracking Algorithm for Autonomous Vehicles in Irregular Road Scenarios Based on Improved CNN

With the continuous development of autonomous driving technology, unmanned vehicles based on object detection algorithms have also begun to receive more and more research and attention. The object detection algorithm based on Convolutional Neural Network (CNN) can accurately detect targets, but its tracking performance in irregular road scenes is not ideal. This article adopts a detection algorithm based on YOLOv3 network, using non-maximum suppression and multi-scale feature fusion technology to solve the target tracking problem in irregular road scenes, and improves the convolutional neural network using YOLOv3 network. The experimental results show that the improved method proposed in this paper can achieve a target detection performance of 96.6 (total of 100) under strong lighting while ensuring detection accuracy. This can effectively improve the target tracking performance in irregular road scenes and cope with situations such as occlusion, deformation, and lighting changes.