Following rapidly and precisely a moving target has become the core functionality in robotic systems for transportation, manufacturing, and medical devices. Among existing targeting following methods, vision-based tracking continues to thrive as one of the most popular, and is the closest method to human perception. However, the low sampling rate and the time delays of visual outputs fundamentally hinder real-time applications. In this paper, we show the potential of significant performance gain in vision-based target following when partial knowledge of the target dynamics is available. Specifically, we propose a new framework with Kalman filters and multi-rate model-based prediction (1) to reconstruct fast-sampled 3D target position and velocity data, and (2) to compensate the time delay. Along the path, we study the impact of slow sampling and the delay duration, and we experimentally verify different algorithms with a robot manipulator equipped with an eye-in-hand camera. The results show that our approach can achieve 95% error reduction rate compared with the commonly used visual servoing technique, when the target is moving with high speed and the visual measurements are slow and incapable of providing timely information.

快速准确地跟踪移动目标已成为运输，制造和医疗设备的机器人系统的核心功能。在现有的针对目标的跟踪方法中，基于视觉的跟踪一直是最流行的方法之一，并且是与人类感知最接近的方法。但是，低采样率和视觉输出的时间延迟从根本上阻碍了实时应用。在本文中，当目标动力学的部分知识可用时，我们将展示在基于视觉的目标中显着提高性能的潜力。具体来说，我们提出了一个具有卡尔曼滤波器和基于多速率模型的预测的新框架（1）重建快速采样的3D目标位置和速度数据，以及（2）补偿时间延迟。沿着这条路 我们研究了慢速采样和延迟持续时间的影响，并通过配备了手持摄像头的机器人操纵器实验性地验证了不同的算法。结果表明，与传统的视觉伺服技术相比，当目标高速运动且视觉测量缓慢且无法及时提供信息时，我们的方法可以实现95％的误差减少率。