Single-view obstacle detection for smart back-up camera systems

This paper presents an implementation of a 3D reconstruction algorithm for the detection of static obstacles from a single rear view parking camera. To guarantee efficiency and accuracy of the solution, we adopted a feature-based approach in which interest points are tracked to estimate the vehicle's motion and multiview triangulation is performed to reconstruct the scene. A full implementation of the algorithm has been achieved on a parallel SIMD array processor unit embedded in a smart automotive camera system. Current in-vehicle beta trials suggest that system's performance meets industrial requirements for real-world use in back-up camera systems.