Obstacle Detection Based on Logistic Regression in Unstructured Environment

Abstract

Obstacles in off-road environments can pose a greater risk to autonomous vehicles, so it is necessary to accurately detect obstacles. This paper proposes an obstacle detection method based on logistic regression. In order to extract the obstacle features better, we first project the discrete point cloud data into the two-dimensional depth map, and then we extract the height difference value and distance difference value between the pixels neighborhoods, after that we use the logistic regression to train and get the corresponding parameters. Combining the training parameters and the extracted effective features, we can obtain the passable probability in the depth map coordinates, and then back-project the depth map pixels into the two-dimensional grid map to obtain the final passable region result. We conduct a number of experiments and the results demonstrate the effectiveness of our method. Furthermore, our method meets the requirements of real-time applications and provides accurate environmental information for unmanned vehicle decision-making and planning.