Comparative Study of Visual Odometry Performance Based on Road Classifications

Dayang Nur Salmi Dharmiza, Awang Salleh and Kuryati, Kipli (2022) Comparative Study of Visual Odometry Performance Based on Road Classifications. Journal of Computer Science, 18 (11). pp. 1030-1037. ISSN 1552-6607

PDF Comparative Study of - Copy.pdf Download (44kB)

Abstract

Accuracy and robustness are among the main concerns in vehicle positioning systems and autonomous applications. These concerns are crucial in GNSS-denied environments; thus, we need an alternative technology to overcome this problem. In recent years, vision-based localization known as visual odometry has gained considerable attention among researchers. Visual odometry is a vision-based pose estimation and it has been developed for mobile object localization such as robots and vehicles while perceiving their environment. Within the last decade, researchers have been immersed in developing techniques to achieve highly accurate and precise localization based on visual odometry. The visual odometry performances are evaluated using an online dataset for benchmarking. Based on the benchmarking, this study reviews and compares the robustness of the recent visual odometry techniques for application, especially in vehicle localization in various road conditions. Evaluation methods for the selected techniques are presented and a thorough analysis of each driving sequence is conducted. The analysis shows that for all visual odometry techniques, localization for high-speed drive suffers higher translation error even though the surrounding has less image noise. Despite that, visual odometry that implements careful feature Selection and Tracking (SOFT) proves to be more robust compared with other techniques with 0.7% relative translation error and a relative rotation error of 0.2 deg/hm.

Actions (For repository members only: login required)

View Item