VECtor: A Versatile Event-Centric Benchmark for Multi-Sensor SLAM (2207.01404v1)

Abstract: Event cameras have recently gained in popularity as they hold strong potential to complement regular cameras in situations of high dynamics or challenging illumination. An important problem that may benefit from the addition of an event camera is given by Simultaneous Localization And Mapping (SLAM). However, in order to ensure progress on event-inclusive multi-sensor SLAM, novel benchmark sequences are needed. Our contribution is the first complete set of benchmark datasets captured with a multi-sensor setup containing an event-based stereo camera, a regular stereo camera, multiple depth sensors, and an inertial measurement unit. The setup is fully hardware-synchronized and underwent accurate extrinsic calibration. All sequences come with ground truth data captured by highly accurate external reference devices such as a motion capture system. Individual sequences include both small and large-scale environments, and cover the specific challenges targeted by dynamic vision sensors.

• The paper introduces VECtor, a novel benchmark dataset featuring hardware-synchronized multi-sensor data, including event, stereo, RGB-D, LiDAR, and IMU, for advancing research in multi-sensor SLAM.
• VECtor includes sequences captured in both small-scale environments with motion capture ground truth and large-scale areas validated with LiDAR-based methods, allowing for comprehensive evaluation of diverse SLAM algorithms.
• This benchmark facilitates the development of robust SLAM systems by providing high-fidelity data for integrating event cameras and traditional sensors, supporting innovation for applications like autonomous navigation.

VECtor: A Comprehensive Event-Centric Benchmark for Multi-Sensor SLAM

The paper "VECtor: A Versatile Event-Centric Benchmark for Multi-Sensor SLAM" by Gao et al. introduces a novel set of benchmark datasets designed to advance research in the integration of dynamic vision sensors (event cameras) into Simultaneous Localization And Mapping (SLAM) systems. This benchmark is a response to the growing interest in event cameras, which offer promising capabilities in environments characterized by high dynamics and challenging illumination conditions.

Contribution and Methodology

The key contribution of this paper is the introduction of a multi-sensor setup that incorporates an event-based stereo camera, a regular stereo camera, an RGB-D sensor, a LiDAR, and an Inertial Measurement Unit (IMU). These sensors are hardware-synchronized and undergo accurate extrinsic calibration to ensure reliable data collection. The dataset includes sequences captured in both small-scale (within a motion capture arena) and large-scale environments (various architectural spaces), providing comprehensive coverage for evaluating SLAM systems.

For small-scale environments, precise six degrees of freedom ground truth is facilitated using a motion capture system. In contrast, for large-scale data sequences, the team employs the Iterative Closest Point (ICP) method to align LiDAR scans with pre-scanned dense point clouds from a FARO laser scanner, providing accurate ground truth.

Results and Evaluation

The datasets are validated using several state-of-the-art SLAM algorithms, including ORB-SLAM3 and VINS-Fusion. The results indicate alignment with the expected performance of these algorithms, underscoring the dataset's suitability for benchmarking purposes. The paper highlights the challenges faced by current event-based SLAM methods, emphasizing the need for further development in this area.

Table comparisons within the paper demonstrate the comprehensive capabilities of the VECtor dataset relative to existing datasets, particularly in terms of synchronization accuracy, sensor diversity, and fidelity of ground truth data. This is illustrated through meticulous evaluation metrics such as Relative Pose Error (RPE) and Absolute Trajectory Error (ATE), which are crucial for assessing both local tracking accuracy and global consistency.

Implications and Future Work

The introduction of VECtor has significant implications for the development of robust SLAM systems, particularly those leveraging the unique attributes of event cameras. By providing high-fidelity, multi-modal data, this benchmark supports the design, testing, and validation of algorithms that can effectively integrate event-based data with traditional sensory inputs.

Theoretical advancements are likely to arise from the paper of event-driven processing, an efficient paradigm supported by the nature of event camera data. Practically, more reliable SLAM algorithms can enhance the capabilities of autonomous systems, such as drones and autonomous vehicles, in dynamic and visually challenging environments.

The authors' future work involves establishing an open SLAM benchmarking platform that facilitates algorithm submissions and evaluations using unpublished datasets, ensuring a fair assessment process and encouraging continuous improvement in SLAM technologies.

In summary, VECtor represents a meaningful step forward in the development of SLAM systems and provides a valuable resource for researchers examining the integration of diverse sensor modalities. The precise hardware synchronization and extensive calibration efforts make it a potent tool for fostering innovation in event-centric multi-sensor fusion.