Obstacle detection using stereovision for Android-based mobile devices

The increasing development of smart mobile devices has brought a new solution to improving driving assistance systems. In this paper, we present an approach for a real-time obstacle detection using stereovision on mobile devices. We started from the premises that regions in the image belonging to the same object have similar appearance and are close in the 3D space. The limitations introduced by the low-computational hardware of the mobile devices imposed the use of a sparse stereo reconstruction, followed by an efficient implementation of superpixel segmentation. The simple linear iterative clustering (SLIC) is well-suited for such applications extracting superpixels with fine precision. We detect the road and markings in order to remove them from the region of interest where potential obstacles are searched. In a further step, the 3D information obtained from stereo reconstruction allows the superpixels to be grouped in regions which belong to the same object. In conclusion, we managed to obtain accurate obstacle detection at short-medium distances and in low-speed traffic scenarios and most importantly real-time processing.