<p>To improve the reliability and safety of the auto drive system in complex scenes, a lane detection algorithm is proposed. This algorithm improves detection performance by stacking features layer by layer, modifying classification methods, and optimizing loss functions. The dynamic object detection and tracking algorithm is based on You Only Look Once version 5 and the deep simple online real-time tracker algorithm. It introduces efficient channel attention in the backbone network, partial convolution modules in the neck network, and optimizes the feature extraction network for improvement, and proposes a collaborative optimization method for lane and object tracking. The results show that the lane detection algorithm has an accuracy of 87.8%, a recall rate of 86.3%, an mIoU of 77.8%, and a processing speed of 46.5fps on the CULane dataset. The improved dynamic object detection algorithm has an accuracy rate of 88.5%.The improved dynamic target tracking algorithm has an average multi-target tracking accuracy of 83.7% and an average multi-target tracking accuracy of 85.3%.After collaborative optimization, the accuracy of lane detection is significantly improved under the influence of dynamic targets, the accuracy of target tracking is improved under lane constraints, and the overall operating efficiency of the system is also improved. Research shows that the proposed algorithm and collaborative optimization method perform well in intricate road situations and ever-changing surroundings.</p>