Enhancing Depth Perception Using Angled Stereo Fisheye Cameras with Egomotion

The realm of vision systems leveraging fisheye cameras has encountered persistent challenges in stereo vision and depth perception, particularly due to lens distortions. This paper addresses the limitations of traditional depth estimation techniques by introducing an innovative approach that com-bines angled stereo fisheye cameras with egomotion dynamics. Utilizing 360-degree cameras, such as the Ricoh Theta series, our methodology enhances depth perception in the peripheral areas of the field of view-regions typically underrepresented in depth data. By taking advantage of the camera's egomotion, we apply geometric principles to compute depth from motion parallax within these peripheral segments without the need for resource-intensive unwarping procedures. Our empirical research, which exploits frames captured at different times to extract depth information, demonstrates over 90% accuracy for target depth estimation in selected peripheral regions. This technique not only reduces computational overhead but also enables a more comprehensive capture of depth data, which has been validated against ground truth measurements from satel-lite imagery. Our findings indicate a promising advancement in spatial awareness for various practical applications, including enhanced navigational tools for both military intelligence and autonomous vehicles. Fisheye lens distortions, traditionally seen as a handicap, are repurposed to effectively calculate depth using the observer's motion. The paper outlines the potential of this technology in improving environmental interaction and navigation, suggesting its transformative capability for industries that rely heavily on spatial accuracy. Future work will integrate optical flow and visual inertial odometry with angled stereo fisheye cameras to improve the accuracy of ego-motion estimations, design seamless transition models between stereo and peripheral vision, and develop robust tracking and detection technologies to analyze complex visual information with enriched depth mapping for broader applications.