A low-cost and robust optical flow CMOS camera for velocity estimation

This paper presents a robust velocity estimation algorithm applicable to low-cost monocular platform with a tri-axial gyroscope. The algorithm is developed based on PX4FLOW platform which is an open software and open hardware solution for monocular speed estimation. The paper provides three main contributions. The first is the improved robustness achieved by various methods including feature block selection, feed-forward rotation, motion compensation, failure case detection and filtering. The second contribution is the improved performance in terms of resolution and search range achieved by adaptive frame interval and advanced search algorithm. The final contribution is a systematical strategy for parameter selection to fully utilize the hardware resources and optimize the performance in different contexts. The performance was benchmarked against the PX4FLOW and GPS. Results indicated that the proposed algorithm achieved levels of robustness exceeding that of PX4FLOW in measuring velocity. Also, the integrated trajectory agreed with the position calculated by an on-board GPS system.