triCAM: A Real Monocular Multi-Modal Event-based Pedestrian Dataset

Alimatou sadia Memudu, Crescenzo Edoardo Mauriello, Yuta Nakano, Jean Martinet

Keywords: Multimodal learning, Multisensor data, Event camera, RGBD camera, Depth estimation, IMU

Abstract: Event-based visions offer key advantages, such as low latency, high dynamic range, and microsecond temporal resolution. These strengths have motivated extensive research into their complementarity with other modalities, which led to the creation of several multi-modal event-based datasets. However, most of these datasets are designed for automotive or robotic domains, with limited attention to human-centered perception in everyday settings. In this paper, we introduce triCAM, a real-world monocular multi-modal event-based pedestrian dataset. triCAM integrates event streams, RGB images, depth images, IMU data, and pedestrian bounding box annotations. This dataset contains 20 sequences, each recorded in two different restaurants in both static and dynamic camera motions. By providing a rich dataset on pedestrian activities in socially interactive environments, triCAM contributes to the advancement of research in robust perception and human interaction understanding.

Primary Area: datasets and benchmarks

Submission Number: 23875