Real-time high speed motion prediction using fast aperture-robust\n event-driven visual flow

Abstract

Optical flow is a crucial component of the feature space for early visual\nprocessing of dynamic scenes especially in new applications such as\nself-driving vehicles, drones and autonomous robots. The dynamic vision sensors\nare well suited for such applications because of their asynchronous, sparse and\ntemporally precise representation of the visual dynamics. Many algorithms\nproposed for computing visual flow for these sensors suffer from the aperture\nproblem as the direction of the estimated flow is governed by the curvature of\nthe object rather than the true motion direction. Some methods that do overcome\nthis problem by temporal windowing under-utilize the true precise temporal\nnature of the dynamic sensors. In this paper, we propose a novel multi-scale\nplane fitting based visual flow algorithm that is robust to the aperture\nproblem and also computationally fast and efficient. Our algorithm performs\nwell in many scenarios ranging from fixed camera recording simple geometric\nshapes to real world scenarios such as camera mounted on a moving car and can\nsuccessfully perform event-by-event motion estimation of objects in the scene\nto allow for predictions of upto 500 ms i.e. equivalent to 10 to 25 frames with\ntraditional cameras.\n