Nonlinear Distortion Calibration of an Optical Flow Sensor for Monocular Visual Odometry

Abstract

With increasing interest in robotics and automation, there exist a wide range of work on visual odometry. Most literature while capable of showing the state-of-the-art, are inaccessible to engineers outside the domain of computer vision. Some companies minimize this gap by providing these advanced algorithms in an integrated solution, however these solutions do not provide the level of accuracy desired as a single sensor solution. By breaking an assumption in the pinhole model, the paper has shown that it is able to increase the accuracy of a commercially available optical flow sensor by more than tenfold. The sensor was calibrated using a high precision magnetic linear actuator without requiring access to the image frame. Performance test conducted in various environments such as canals and corridors show that accurate visual odometry can be achieved with a small and lightweight commercially available sensor.