Vision-based Coordinate Transformation with Back Propagation Neural Networks on Mobile Robots

Target tracking is important for vision-based robots to implement tasks of grasping, assembling and avoiding obstacles. The purpose of a target tracking system is to identify a target and then to estimate the position of the target. The targets' positions are usually described by various coordinate systems for different purposes. This study focuses on the problem of coordinate transformation on mobile robots and employs the techniques of Back-Propagation Neural Networks to discover the prediction models. With such prediction models, coordinate transformation can be done with less processing time. The techniques have been implemented and integrated with a four-wheeled vision-based security robot and has been verified in real environments. The experimental results show that the proposed method is able to produce simple and precise transformation models and improves the robot's performances.