Canonical Trinocular Camera Setups and Fisheye View for Enhanced Feature-Based Visual Aerial Odometry

This comprehensive study delves into the intricacies of trinocular camera configurations for feature-based Visual Odometry (VO) in aerial robotics, emphasizing a comparative analysis of the proposed trinocular Stereo Feature Matching (SFM) methodologies in terms of computational efficiency and accuracy against the established binocular SFM by comparing their VO results. It focuses on two canonical trinocular camera setups: in-line and 45-degree, each incorporating a rectilinear or 180-degree field-of-view fisheye lens for the third camera. The in-line configuration positions an additional camera at the center of a binocular setup, facilitating a systematic coarse-to-fine search in stereo feature matching. The 45-degree setup introduces a triangular camera arrangement, altering the epipolar line's gradient from 0 to 45 degrees to enhance the reliability of stereo feature matching during a coarse-to-fine search similar to that of the in-line configuration. The proposed VO methodology is rigorously tested within a simulated quadrotor flight environment, created explicitly for this study using the Unity game engine to replicate realistic outdoor navigation scenarios. The simulation features a dynamically accurate quadrotor model to generate synchronized trinocular image sets, providing a robust platform for evaluating the VO approaches. The findings highlight that the 45-degree setup with a rectilinear lens for the third camera improves the binocular SFM for VO and decreases the computational cost to less than half, albeit at the expense of obtaining and storing an additional image at each instance.