Real-Time Path Planning and Control for Intelligent Vehicles' Inertial Navigation Based on Computer Vision

Abstract

In order to improve the positioning accuracy and path planning efficiency of intelligent vehicles in dynamic environments, this paper proposes a real-time path planning and control method based on the fusion of computer vision and inertial navigation. The system adopts a three-layer hierarchical architecture that integrates a stereo camera, a six-axis IMU inertial measurement unit, and encoder data to achieve accurate positioning and path planning. The core hardware platform is chosen to be NVIDIA Jetson AGX Xavier, which supports efficient parallel computing and can handle complex vision and navigation algorithms. To enhance the real-time and robustness of the system, the system uses an improved ORB-SLAM3 vision algorithm to construct an environment map through feature extraction and matching, while fusing IMU data to improve positioning stability. In addition, path planning uses the improved Hybrid A* algorithm, which takes vehicle kinematic constraints into account to optimise path search efficiency and planning smoothness. In terms of control, the system achieves optimisation of path tracking by means of a multi-layer control structure, which utilises the joint regulation of model predictive control (MPC) and PID controller to improve path tracking accuracy and response speed. The experimental results show that the system is able to achieve efficient and stable path planning in complex environments, with the positioning error controlled within 5 cm and the path planning frequency reaching 60 Hz, demonstrating excellent real-time performance and robustness.