Research on Unstructured Road Boundary Detection

For the difficult problem that unstructured roads are complex and diverse, with strong interference in the surrounding environment and many uncertainties such as road shadows, water traces and light changes, it is difficult to form a universal road boundary detection, this paper proposes a boundary detection method applicable to unstructured roads. The road boundary detection average error under unstructured roads is within 20cm by generating adversarial networks for road shadow removal, adaptive threshold adjustment under HSV channel, acquiring lane boundaries based on the maximum convex polygon in the connected domain, and using KCF tracking algorithm for anomaly detection rejection.