<title>Adaptive tracking of objects for a mobile robot using range images</title>

Abstract

This paper presents a method for adaptively tracking an object given a sequence of range images for a free flying space robot. The tracker assigns unique identifiers to each new object that it detects, updates the object position and velocity information each frame, feeds and later utilizes position information from a Translational State Estimator, and resolves ambiguities due to object occlusion. The tracker must also remove objects from the world model, as well as successfully identify body parts in images. The tracker uses rough object position and size information to correspond objects frame to frame until the Translational State Estimator (TSE) has enough data to provide accurate position information. Objects that are occluded by other objects are over-segmented and individual sub-blobs are assigned to their respective objects based on proximity, or by correspondence with previously identified sub-blobs. Relative range, TSE accuracy, and priority information is maintained for each object and analysis varies depending on these properties. The proposed system has been extensively tested on simulated range images using a simulator for the EVA Retriever robot.