Research on Multi-Sensor Fusion Embedded Omnidirectional Vision Robot Based on RT1064*

Abstract

Aiming at inherent limitations of single-vision perception, insufficient embedded computing capacity and low tracking precision of miniature omnidirectional robots, this paper develops a multi-sensor fusion embedded vision robot based on NXP RT1064 microcontroller. A dual independent two-stage LDO(Low Dropout Regulator) power circuit is designed to eliminate module crosstalk. A four-layer fusion framework covering data, feature, decision and control layers is constructed to support switchable line-tracking and target recognition modes. Two lightweight neural networks are deployed on distributed vision modules for real-time target detection and classification, with dual consecutive verification to enhance anti-interference performance. The inverse kinematic model of 120° omnidirectional Y-shaped chassis is established, and a cascade PID control strategy is adopted to improve vehicle motion stability. Practical tests verify that the maximum tracking error is limited within $\pm \mathbf{5 ~ c m}$, the stable moving speed reaches $\mathbf{1 . 5 ~ m} / \mathbf{s}$, target detection accuracy is 97.5% and pattern recognition accuracy hits 98%. The modular low-cost system provides a feasible engineering reference for small embedded mobile robots.