Active Event Alignment for Monocular Distance Estimation

TL;DR

This behavior-driven method mimics how biological systems, like the human eye, stabilize their view based on ob-ject distance: distant objects require minimal compensatory rotation to stay in focus, while nearby objects demand greater adjustments to maintain alignment.

Abstract

Event cameras provide a natural and data efficient representation of visual information, motivating novel computational strategies towards extracting visual information. Inspired by the biological vision system, we propose a behavior driven approach for object-wise distance estimation from event camera data. This behavior-driven method mimics how biological systems, like the human eye, stabilize their view based on object distance: distant objects require minimal compensatory rotation to stay in focus, while nearby objects demand greater adjustments to maintain alignment. This adaptive strategy leverages natural stabilization behaviors to estimate relative distances effectively. Unlike traditional vision algorithms that estimate depth across the entire image, our approach targets local depth estimation within a specific region of interest. By aligning events within a small region, we estimate the angular velocity required to stabilize the image motion. We demonstrate that, under certain assumptions, the compensatory rotational flow is inversely proportional to the object's distance. The proposed approach achieves new state-of-the-art accuracy in distance estimation - a performance gain of 16% on EVIMO2. EVIMO2 event sequences comprise complex camera motion and substantial variance in depth of static real world scenes.

Figures (7)

• Figure 1: Akin to gaze stabilization, active event alignment stabilizes a local image region by applying a rotation that counteracts the camera's motion. This rotation leads to locally well aligned events as pictured in (d). The relative distance between two objects can then be inferred by comparing their compensatory rotations.
• Figure 2: Overview of our algorithm. We estimate relative object-wise-distance from active event alignment. Given a set of events, we process alignment in a object-wise fashion. Object regions maybe determined by a provided object segmentation mask or a default segmentation mask (e.g honeycomb) without semantic information. The obtained relation between different angular velocities within the image plane determine the respective object-wise relative depth.
• Figure 3: Qualitative results of object-wise relative depth estimation over time. Top: The line plot shows the relative depth estimation of the object with ID 24 (highlighted with a green circle) of the event sequence scene_03_00_000000.
• Figure 4: Qualitative results of depth estimation on the EVIMO2 dataset, with framewise results on four exemplary video sequences: (a) Events and segmentation mask. (b) Ground truth. (c) E2Depth hidalgo2020learning. (d) EMVS rebecq2018emvs. (e) Ours.
• Figure 5: Qualitative evaluation of region-wise distance estimation w/o object masks. We use a honeycomb grid to define pixel regions for depth estimation. Relative distances and confidence maps are shown in grayscale (white = low confidence, black = high confidence). (a) Events and segmentation masks. (b) Original ground truth. (c) Ground truth using honeycomb regions. (d) Our method using honeycomb regions. (e) $\pm 3\sigma$ confidence interval.