Design and Algorithm Optimization of a Vision-Ultrasound Collaborative Embedded System for Precision Machining

To address the challenges of poor real-time performance and low collaboration efficiency between visual perception and execution control in precision machining, this paper proposes a lightweight vision-ultrasound collaborative embedded system. A fast 3D reconstruction algorithm based on three views is designed. By integrating an improved Canny edge detection method and a fast ICP registration algorithm, the positioning time is reduced to 0.8 seconds. A dynamic regulation model for ultrasonic parameters based on reinforcement learning is proposed, achieving millisecond-level decision-making response. The system is deployed on the Jetson Orin edge computing platform. Through model pruning and quantization, the inference latency is measured at 30.30 ms on the embedded platform, meeting the real-time requirement. Experimental results demonstrate that the system achieves real-time closed-loop collaboration between vision and ultrasound while maintaining high-precision positioning, providing an efficient embedded solution for intelligent machining.