Passable Area Detection in Off-road Environments Based on Improved PSPNet

In order to improve the adaptability of autonomous vehicles in off-road environments, detecting passable areas is a necessary step that cannot be avoided. Compared to structured roads in urban environments, road extraction in off-road environments is more complex. This paper conducts research on various off-road scenarios and proposes an improved method for passable area extraction using a modified PSPNet. During feature extraction, a lightweight MobileNetv2 network is utilized with depthwise separable convolution and residual edge modules. In addition, dilation convolution modules are added and shallow image features are fused. Experimental results demonstrate that the proposed method achieves better performance in passable area detection in off-road environments, and has high application value and practicality.