A parallel implementation of a multisensor feature-based range-estimation method

There are many proposed vision based methods to perform obstacle detection and avoidance for autonomous or semi-autonomous vehicles. A system capable of supporting autonomous helicopter navigation will need to extract obstacle information from imagery at rates varying from ten images per second to thirty or more images per second depending on the vehicle speed. This paper describes an efficient and flexible parallel implementation of a multisensor feature-based range-estimation algorithm, targeted for automated helicopter flight. The algorithm can track hundreds of features in multiple image sensors using an extended Kalman filter to estimate the feature´s location in a master sensor coordinate frame. The feature-tracking algorithm has reached relative maturity in the laboratory and is now being ported to several real-time architectures to support autonomous helicopter navigation research, The focus of this paper is not the core theory of the vision algorithm itself, but those aspects of it that affect the method of parallelization. The performance of the parallel algorithm is analyzed, with respect to three load balancing schemes, on both a distributed-memory and shared-memory parallel computer