Spherical formulation of moving object geometric constraints for monocular fisheye cameras

Abstract

In this paper, we introduce a moving object detection algorithm for fisheye cameras used in autonomous driving. We reformulate the three commonly used constraints in rectilinear images (epipolar, positive depth and positive height constraints) to spherical coordinates which is invariant to specific camera configuration once the calibration is known. One of the main challenging use case in autonomous driving is to detect parallel moving objects which suffer from motionparallax ambiguity. To alleviate this, we formulate an additional fourth constraint, called the anti-parallel constraint, which aids the detection of objects with motion that mirrors the egovehicle possible. We analyze the proposed algorithm in different scenarios and demonstrate that it works effectively operating directly on fisheye images.

Spherical Formulation of Geometric Motion Segmentation Constraints in Fisheye Cameras

“…In the geometrical treatment of moving object detection, the use of spherical coordinates has been investigated before, especially with different formulations of the reasonably well known epipolar constraint [9], [10]. For a more comprehensive review of these methods, we refer to [3]. In particular, Marković et al [11] propose the closest to one of our approaches, describing closely the spherical epipolar constraint ( §III-B), though this is not explicitly mentioned.…”

“…through visual odometry or kinematics available on vehicle system bus). A summary of the described constraints is presented in Table 1 of [3].…”

“…This is the source of most of the false positives in Table I. The work presented in [3] contains a deeper analysis of this figure and of the visual performance of the constraints.…”

“…We structure the paper in the following way. In Section II, we augment the background material from [3] with some additional material. In Section II-B, fisheye mapping is described so that the reader can understand the required mapping from fisheye image space to spherical coordinates.…”

“…This work was initially presented in [3], with significant additional material added to this paper, including a much deeper description of the geometric constraints (and describing the additional three-view constraint), adding background material in the odometry and optical flow, and providing additional description of the results, in particular a comparison with FisheyeMODNet [4]. Additional results are presented, for example the detection ranges in Figure 14.…”