Drivable Area and Road Anomaly Segmentation using SSLG with V-Disparity Maps

Abstract

Real world applications like robotic wheelchairs need the automatic detection of roads, potholes, and anomalies. Conventional image processing methods perform the improper recognition of anomalies and result in poor performance. Thus, this article mainly focuses on the implementation of Self-Supervised Label Generator (SSLG) based road anomaly detection system using vertical disparity maps. Initially, the disparity maps are used to identify the borders of the road and then anomalies are identified using filtered disparity maps. Further, the depth anomaly map is calculated using probabilistic approaches. Further, the implementations are performed on real world Red-Green-Blue-Depth (RGB-D) dataset. The simulation results show that the performance of proposed method results in superior performance as compared to the state-of-the-art approaches.