Obstacle detection by stereo vision, introducing the pq method

Safe, robust operation of an autonomous vehicle in cross-country environments relies on sensing of the surroundings. Thanks to the reduced cost of vision hardware, and increasing computational power, computer vision has become an attractive alternative for this task. This paper concentrates on the use of stereo vision for obstacles detection in cross-country environments where the ground surface can not be modeled as ramps, i.e. linear patches. Given a 3D reconstruction of the surrounding environment, obstacles are detected using the concept of compatible points. The concept classify points as obstacles if they fall within the volume of cone located with its apex at the point being evaluated. The cone may be adjusted adjusted according the physical parameters of the vehicle. The paper introduces a novel Projection and Quantification method that based on vehicle orientation rotates the 3D information onto an intuitive two dimensional surface plot and quantifies the information into bins adjusted to the specific application. In this way the search space for compatible points is significantly reduced. The new method is evaluated for a specific robotic application and the results are compared to previous results on a number of typical scenarios. Combined with an intuitive representation of obstacles in a two dimensional surface plot, the results indicate a significant reduction in the computational complexity for relevant scenarios.