Research on Road Traversability Detection Using Edge Computing and Deep Learning

Oriented to the challenges of unstructured environments such as fuzzy road features, variable road types, lack of clear road boundaries, and high real-time maneuvering requirements of unmanned platforms, a passable area detection method based on edge computing and deep learning is proposed. The unmanned platform collects unstructured environmental data, trains the dual-channel convolutional neural network BiSeNetV2, and utilizes offline compilation and online inference to transplant the algorithmic model into the edge computing module, realizing real-time passable area detection on the unmanned platform. Through verification, the method can be adapted to a variety of complex unstructured environments, such as unpaved roads, paved roads, snow, etc., and provides favorable passable area detection results, with a detection accuracy of 84.75% and an average detection time of 89.08 ms, which can provide real-time guidance information for unmanned platform scene modeling, autonomous path planning, and remote-control driving, etc., and contribute to the stable passage of unmanned platforms in unstructured environments.