Detecting parallel-moving objects in the monocular case employing CNN depth maps

Nolang Fanani, Matthias Ochs, Rudolf Mester; Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 2018, pp. 0-0

This paper presents a method for detecting independently moving objects (IMOs) from a monocular camera mounted on a moving car. We use an existing state of the art monocular sparse visual odome-try/SLAM framework, and specifically attack the notorious problem ofidentifying those IMOs which move parallel to the ego-car motion, thatis, in an ’epipolar-conformant’ way. IMO candidate patches are obtained from an existing CNN-based car instance detector. While crossing IMOscan be identified as such by epipolar consistency checks, IMOs that moveparallel to the camera motion are much harder to detect as their epipo-lar conformity allows to misinterpret them as static objects in a wrongdistance. We employ a CNN to provide an appearance-based depth es-timate, and the ambiguity problem can be solved through depth veri-fication. The obtained motion labels (IMO/static) are then propagatedover time using the combination of motion cues and appearance-basedinformation of the IMO candidate patches. We evaluate the performanceof our method on the KITTI dataset.