Abstract
This paper presents a visual odometer system using stereo cameras for pedestrian navigation. A novel method for pedestrian navigation based on the knowledge of gait analysis and robust ego-motion estimation is proposed. Two major problems of implementing the system on a pedestrian are stated. Firstly, the features collected from cameras attached on a walking pedestrian normally have winding trajectory resulting in inaccurate tracking. Secondly, the observed object moving independently leads to incorrect ego-motion estimation. Using gait analysis, capturing images at the same stage of the walking cycle produces a less winding trajectory that allows tracking without stabilizing the images. Robust ego motion is also introduced to eliminate outliers that are independently moving features, mismatched features in the stereo matching step and incorrectly assigned features in the tracking step. Data processing techniques including corner detection, stereo matching, triangulation, tracking, and ego-motion estimation are employed. The outcome is the estimated incremental ego motion of the stereo cameras. The approach not only enables the system to operate on walking users but also improves the accuracy of ego-motion estimation.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have