Visual odometer for pedestrian navigation

This paper presents a visual odometer system using stereo cameras for pedestrian navigation. Corner detection, stereo matching, triangulation, tracking, and robust ego-motion estimation are used for data processing. The outcome is the estimated incremental egomotion of the stereo cameras. The problems of implementing the system on a pedestrian are stated. The first problem is image feature motion. The motion of image features is the result of the motion of stereo cameras. In the case that the feature belongs to an independent moving object, the movement of the feature is the result of the motion of the cameras together with the motion of the feature itself. Hence, a novel robust ego-motion estimation algorithm must be utilized to eliminate outliers, which are independent moving features, mismatched features in the stereo matching step and incorrect assigned features in the tracking step. Secondly, the feature, which is collected on a pedestrian, results in a winding trajectory, which may easily fail the tracking algorithm. In this paper, we introduce a new method based on the knowledge of gait analysis to capture images at the same stage of walking cycle. This leads to less winding trajectory, which can be tracked without increasing order and computational cost of the tracker.