基于原始事件流的事件相机/IMU时空标定

Spatiotemporal calibration of event camera/IMU systems based on raw event streams

关键词:  惯性测量单元  事件相机  时空标定  对比度优化  原始事件流

Key words:  Inertial measurement unit(IMU)  Event camera  Spatiotemporal calibration  Contrast optimization  Raw event streams

摘要:

近年来，相机/IMU传感器组合因其轻量化设计与紧凑结构，在机器人、增强现实(AR)及自动驾驶领域得到了广泛应用。然而，传统的标准相机存在感知延迟和低动态范围等固有缺陷，促使新型事件相机的引入以提升系统的完备性和可靠性。事件相机/IMU标定作为鲁棒运动估计的前提，现有方法通常依赖复原的标准图像与人工靶标，导致流程繁琐且需要人工干预。针对此问题，提出了一种基于对比度最大化的无靶标事件相机/IMU旋转外参和时延标定方法。该方法直接利用原始异步事件流，无需复杂的手动初始化或先验知识。具体而言，首先设计了滑动窗口机制估计短时段内事件相机的角速度。在此基础上，基于连续时间表示，联合优化IMU陀螺仪因子与事件相机角速度因子，以求解最优旋转参数和时间延迟参数。在公开与自采数据集上的实验表明，该方法可以实现与基于靶标的先进标定方法相当的精度。

Abstract:

In recent years, camera/IMU sensor suites have been widely adopted in robotics, augmented reality (AR), and autonomous driving due to their lightweight design and compact structure. However, conventional standard cameras have inherent limitations, such as perception latency and a low dynamic range, prompting the integration of novel event cameras to enhance system integrity and reliability. Current event camera/IMU calibration methods for robust motion estimation typically rely on reconstructed standard images and artificial targets, resulting in complicated procedures that require manual intervention. To address this problem, a target-free method for calibrating the rotational extrinsic parameters and time delay of event camera/IMU systems based on contrast maximisation is proposed. This method directly processes raw asynchronous event streams without complex manual initialization or prior knowledge. Specifically, a sliding window mechanism is designed to estimate the short-term angular velocities of the event camera. Based on this, a continuous-time representation is employed to jointly optimize the IMU gyroscope factors and the event camera angular velocity factors for determining the optimal rotation and time delay parameters. Experiments conducted on both public and self-collected datasets demonstrate that the proposed method achieves accuracy comparable to state-of-the-art target-based calibration approaches.