With the path-tracking advancement of smart logistics and intelligent manufacturing technologies, the demand for autonomous navigation of AGVs (Autonomous Guided Vehicles) in complex environments is continuously growing. To address the high-precision navigation requirements of AGVs in dynamic environments, this paper proposes a hybrid navigation scheme that integrates laser SLAM-based global navigation with vision-based local path tracking. Additionally, a deep learning-based obstacle detection and localization module based on YOLOv5 is integrated into the system. The system first employs a modified Cartographer algorithm to construct a high-precision 2D grid map and utilizes the Adaptive Monte Carlo Localization (AMCL) algorithm for real-time AGV localization. In terms of navigation strategy, the system achieves coordinated optimization of global path planning and local dynamic obstacle avoidance. It further integrates the YOLOv5 vision detection module to identify, classify, and localize dynamic obstacles, while combining a visual neuron-based PID controller to enhance path-tracking accuracy. Experimental results demonstrate that the proposed method exhibits strong performance in obstacle detection and effectively improves the operational reliability and precision of AGVs in complex environments.