Benefit of large field-of-view cameras for visual odometry

Abstract

The transition of visual-odometry technology from research demonstrators to commercial applications naturally raises the question: "what is the optimal camera for vision-based motion estimation?" This question is crucial as the choice of camera has a tremendous impact on the robustness and accuracy of the employed visual odometry algorithm. While many properties of a camera (e.g. resolution, frame-rate, global-shutter/rolling-shutter) could be considered, in this work we focus on evaluating the impact of the camera field-of-view (FoV) and optics (i.e., fisheye or catadioptric) on the quality of the motion estimate. Since the motion-estimation performance depends highly on the geometry of the scene and the motion of the camera, we analyze two common operational environments in mobile robotics: an urban environment and an indoor scene. To confirm the theoretical observations, we implement a state-of-the-art VO pipeline that works with large FoV fisheye and catadioptric cameras. We evaluate the proposed VO pipeline in both synthetic and real experiments. The experiments point out that it is advantageous to use a large FoV camera (e.g., fisheye or catadioptric) for indoor scenes and a smaller FoV for urban canyon environments. Benefit of Large Field-of-View Cameras for Visual OdometryZichao Zhang, Henri Rebecq, Christian Forster, Davide Scaramuzza Abstract-The transition of visual-odometry technology from research demonstrators to commercial applications naturally raises the question: "what is the optimal camera for visionbased motion estimation?" This question is crucial as the choice of camera has a tremendous impact on the robustness and accuracy of the employed visual odometry algorithm. While many properties of a camera (e.g. resolution, frame-rate, globalshutter/rolling-shutter) could be considered, in this work we focus on evaluating the impact of the camera field-of-view (FoV) and optics (i.e., fisheye or catadioptric) on the quality of the motion estimate. Since the motion-estimation performance depends highly on the geometry of the scene and the motion of the camera, we analyze two common operational environments in mobile robotics: an urban environment and an indoor scene. To confirm the theoretical observations, we implement a stateof-the-art VO pipeline that works with large FoV fisheye and catadioptric cameras. We evaluate the proposed VO pipeline in both synthetic and real experiments. The experiments point out that it is advantageous to use a large FoV camera (e.g., fisheye or catadioptric) for indoor scenes and a smaller FoV for urban canyon environments. SUPPLEMENTARY MATERIALA video showing our omnidirectional visual odometry pipeline performing on real and synthetic data is available at the website: http://rpg.ifi.uzh.ch/fov.html.

Computer Vision Methods for Automotive Applications

“…In this section we present three applications of computer vision methods in the automotive industry and robotics. In order to highlight the potential of the methods, we make use of simulations using Blender, which is an open-source platform used for generating virtual scenarios as ground truth data for evaluation [13], [14]. In Section 3.1, real images of a wheel were used.…”

Affine Correspondences and Challenges in Multi-Sensor Systems

“…First row: Frames (1st row ) from the scene 'vfr' [188]. Further rows: Rotation errors versus the number of inliers and the percentage of outliers for 2AC (second row ) and 5PT (third row ).…”

FAST-FUSION: An Improved Accuracy Omnidirectional Visual Odometry System with Sensor Fusion and GPU Optimization for Embedded Low Cost Hardware

“…Zhang et al propose a version of semi-direct visual odometry (SVO) [16] that also uses wide field of view cameras. They implement a camera model suitable for these cameras and use samples from the epipolar curves present in the omnidirectional VO configuration to estimate the camera motion [7].…”

“…Also, to perceive the motion scale of a mobile robot using a single camera without any prior knowledge about the environment or other sources of information is not possible due to the unavailability of depth information [6]. To improve VO performance, the use of omnidirectional cameras in this context has increased since they allow to capture more information about the scene and to track individual features over a more extensive set of consecutive images [7]. Even so, monocular omnidirectional VO benefits from the use of other sources of information such as inertial and/or range sensors.…”