PixVOD: Pixel-Distributed Direct Visual Odometry and Depth Estimation
Research area: Computer Vision (CV) Authors: Shinjeong Kim, Ignacio Alzugaray, Callum Rhodes, Paul H. J. Kelly, Andrew J. Davison Published: 2026-06-02 arXiv: 2606.03989
Abstract
Images composed of 2D pixel arrays are the standard input to computer vision algorithms, yet many underlying computations can be distributed across pixels. Transmitting raw, redundant, and noisy pixel data off the sensor remains inefficient, motivating a shift toward focal-plane sensor-processors that perform a significant part of the computation directly within each pixel.
The authors envision pixels synthesizing higher-level signals locally, reducing downstream load, and providing richer inputs for higher-level vision tasks. They propose a fully parallelizable form of visual odometry and depth estimation across pixels, where sensor-processors exchange information through Gaussian Belief Propagation (GBP) to achieve consensus about camera motion and infer depth from per-pixel photometric observations combined with surface normal priors.
Key Technical Contributions
- Pixel-distributed computation: Visual odometry and depth estimation are formulated so that computation is distributed across individual pixels rather than performed centrally on raw pixel data.
- Gaussian Belief Propagation (GBP): Pixel-level sensor-processors exchange information via GBP to reach consensus on camera motion.
- Per-pixel depth inference: Depth is inferred from per-pixel photometric observations with surface normal priors.
- Keyframe-like anchoring mechanism: To maintain geometric stability during optimization, an anchoring scheme regulates the effective baseline between frames, enabling consistent motion and depth updates.
*Auto-collected on 2026-06-04*