Depth-data-based object cluster tracking and velocity estimation in robot workspace

Abstract

Depth-data-based sensor systems, such as depth cameras or LiDAR systems, are gaining popularity in the field of robotics, especially in human-robot collaboration. To avoid collisions with humans or external objects, object detection and tracking in the workspace is needed. This paper presents an integrated object cluster tracking and velocity estimation method that is purely based on depth data. Therefore, a tracking heuristic based on similarity and the velocity of the object is used to enable the tracking of external objects. To obtain the velocity, a Kalman filter utilizing a constant velocity model is implemented. For experimental verification, a case study comprising two objects moving within the robot workspace is designed. The experimental setup allows for the initial tracking a maximal trackable object velocity of 9m/s , and for already tracked objects a velocity deviation of 3.4m/s to correctly track both repetitive and arbitrary motions of the test objects, and thus constitutes the proposed integrated object cluster tracking approach as a foundation for collision avoidance strategies in robotic tasks.