Obstacles are one kind of important environment objects for robots and intelligent vehicle systems. It is difficult and unnecessary to construct 3D or image model for every obstacle, because it could appear anywhere and its motion status is uncertain. The common approach towards the detection of obstacles searches for cues which are incompatible with the assumption that a certain area of the image corresponds to a road surface: such cues could be significant gray value transitions, texture boundaries, or -- alternatively -- regions with unexpected gray values or colors. Usually their detection rate is low or their false alarm rate is high, and is weak for the process of shade regions or texture regions.

Unlike the traditional approach (e.g. CMU Navlab), which assumes a certain area of the image corresponds to a road surface, we use one more reliable approach, where we essentially exploit the "recognition by alignment" technique based on "Plane+Parallax" to solve the obstacle detection problem. This tells us that anything extending from the assumed planar road will exhibit a disparity that differs from those points on the road plane itself if image frames recorded from different vantage points are compared.

Knowing the general formula constraining two correspondent points in stereo images by using projective geometry: the Relative Affine Structure, we use a special case: the planar projection stereopsis, which detects any obstacle extending from the road plane.

Using the ground plane as the reference plane, our system can reliably detect obstacle (of size 20x20cm) at 50m. And the whole processing cycle is less than 200ms, which meets the real-time performance need of our THMR-V.

Here are some examples with heavy shadow/texture: