An evaluation of stereo vision for distance estimation using the SGBM algorithm in the CARLA simulator

This paper presents an evaluation of stereo vision based on the semi-global block matching (SGBM) algorithm for distance estimation in an autonomous parking scenario using the CARLA simulator. Distance-disparity regression functions are explored to enhance distance estimation accuracy. The proposed distance estimation model was evaluated using the design science research methodology (DSRM) framework, with experimental validation conducted in CARLA’s promenade environment. The evaluation employed root mean square error (RMSE) and relative error metrics to assess performance. Experiments were performed within a range of 40-350 cm, which is relevant for autonomous parking applications. The experimental results show that the algorithm achieves an overall RMSE of 1.69 cm and an average relative error of 1.1 %. The findings contribute to the advancement of perception systems for autonomous vehicles, particularly in challenging environments.