ECMD: An Event-Centric Multisensory Driving Dataset for SLAM

Abstract

Leveraging multiple sensors enhances complex environmental perception and increases resilience to varying luminance conditions and high-speed motion patterns, achieving precise localization and mapping. This paper proposes, ECMD, an event-centric multisensory dataset containing 81 sequences and covering over 200 km of various challenging driving scenarios including high-speed motion, repetitive scenarios, dynamic objects, etc. ECMD provides data from two sets of stereo event cameras with different resolutions (640×480, 346×260), stereo industrial cameras, an infrared camera, a top-installed mechanical LiDAR with two slanted LiDARs, two consumer-level GNSS receivers, and an onboard IMU. Meanwhile, the ground-truth of the vehicle was obtained using a centimeter-level high-accuracy GNSS-RTK/INS navigation system. All sensors are well-calibrated and temporally synchronized at the hardware level, with recording data simultaneously. We additionally evaluate several state-of-the-art SLAM algorithms for benchmarking visual and LiDAR SLAM and identifying their limitations. The dataset is available at https://arclab-hku.github.io/ecmd/.

Problem Addressed

SLAM research lacks multisensory driving datasets that combine event cameras with conventional sensors across challenging conditions such as varying luminance, high-speed motion, repetitive scenes, and dynamic objects.

Key Contributions

• ECMD dataset with 81 sequences covering over 200 km of challenging driving scenarios
• Rich sensor suite: two stereo event camera sets (640x480 and 346x260), stereo industrial cameras, infrared camera, three LiDARs including two slanted, consumer GNSS receivers, and onboard IMU
• Hardware-level temporal synchronization and careful calibration of all sensors, with centimeter-level GNSS-RTK/INS ground truth
• Benchmark evaluation of state-of-the-art visual and LiDAR SLAM algorithms identifying their limitations

Conclusions

All sensors are meticulously calibrated and hardware-synchronized, with centimeter-level ground truth from a GNSS-RTK/INS system. Benchmarking of state-of-the-art visual and LiDAR SLAM algorithms reveals their constraints, and the dataset is publicly released to support SLAM development, with future expansion planned for semantics, optical flow, and depth estimation.