Target acquisition using Natural Feature Image Recognition

Natural feature image recognition (NFIR) is camera based robotic vision system for recognition, acquisition, tracking and pose estimation of a target vehicle. This paper presents our on-going work on development of the capability of the NFIR software in recognition and acquisition for autonomous rendezvous and docking. The heart of the acquisition algorithm is a classification-based scheme. Feature points with high intensity gradients are selected as key features and point matching is formulated as a multi-class classification problem. The proposed scheme shifts much of the computation burden to a training phase to ensure an efficient and robust run-time performance on target recognition as demonstrated by experimental results.