E-Calib: A Fast, Robust and Accurate Calibration Toolbox for Event Cameras

TL;DR

E-Calib is presented, a novel, fast, robust, and accurate calibration toolbox for event cameras utilizing the asymmetric circle grid, for its robustness to out-of-focus scenes and outperforms the state-of-the-art in detection success rate, reprojection error, and pose estimation accuracy.

Abstract

Event cameras triggered a paradigm shift in the computer vision community delineated by their asynchronous nature, low latency, and high dynamic range. Calibration of event cameras is always essential to account for the sensor intrinsic parameters and for 3D perception. However, conventional image-based calibration techniques are not applicable due to the asynchronous, binary output of the sensor. The current standard for calibrating event cameras relies on either blinking patterns or event-based image reconstruction algorithms. These approaches are difficult to deploy in factory settings and are affected by noise and artifacts degrading the calibration performance. To bridge these limitations, we present E-Calib, a novel, fast, robust, and accurate calibration toolbox for event cameras utilizing the asymmetric circle grid, for its robustness to out-of-focus scenes. The proposed method is tested in a variety of rigorous experiments for different event camera models, on circle grids with different geometric properties, and under challenging illumination conditions. The results show that our approach outperforms the state-of-the-art in detection success rate, reprojection error, and estimation accuracy of extrinsic parameters.

Figures (15)

• Figure 1: A high-level block diagram of the proposed calibration method. First, the raw events $\mathbf{E}_{k}$ are acquired from the event camera and are clustered to their respective circles to the clustered event sets $C = \{ {I_1, I_2, ... I_{J}} \}$ using ST-DBSCAN. Second, the centers of the circles are robustly extracted by means of eRWLS and the pattern is distinguished from the background using modified hierarchical clustering. Finally, the calibration optimization is performed using the predicted circles 2D image points $\mathbf{U}_{M}$ to obtain the event camera intrinsics.
• Figure 2: Events generated for an event camera observing a) asymmetric circles pattern and b) checkerboard. Notice that the checkerboard edges parallel to camera motion do not fire events, while events for circular features are motion invariant.
• Figure 3: Accumulated spatiotemporal window for an event camera observing the circles pattern. Notice that the circle targets form dense slanted cylinders due to the monotonic increase in time.
• Figure 4: a), b) and c), d) ST-DBSCAN on $e_{k}$ and $\hat{e}_{k}$ for DAVIS346 and DVXplorer, respectively. A fixed $\epsilon_{s}$ on $\hat{\mathbf{E}}_{k}$ maintains the desired clustering performance.
• Figure 5: The clustered events from ST-DBSCAN. Noisy events (outliers) are still evident close to the observed circle targets. Note that the figure is for illustrative purposes and events extend in the time dimension.