Stereovision on mobile devices for obstacle detection in low speed traffic scenarios

Since smart mobile devices having capabilities of synchronous stereo image acquisition have been released on the market, the topic of real-time 3D environment reconstruction by stereovision on such mobile platforms has become of a greater interest among researchers. In this paper we continue the sparse stereovision approach proposed in [15], while focusing on improving the reconstruction results by refining the disparity computation accuracy to a sub-pixel level and by using the available sensors to gain more information about the position of the device relative to the world. After the 3D points are reconstructed by triangulation, a correction is applied on them to compensate for a possible pitch rotation of the device. Moreover, we present a fast approach for detecting the obstacle on the estimated trajectory of the vehicle. A series of experiments have been conducted which proved that although mobile development is constrained by the available features of the device and its operating system, sensor information is beneficial, and more importantly, both reconstruction accuracy and obstacle detection at short-medium distances and real-time processing can be achieved. Thus, developing driving assistance functions with such devices is possible for low vehicle speeds / short range scenarios, which often occur in urban environments.