FEMOT: Multi-Object Tracking using Frame and Event Cameras

Researchers have proposed new algorithms and datasets to improve multi-object tracking and pose estimation, addressing limitations in current methods.

A team of researchers has introduced FEMOT, a large-scale RGB-event multi-object tracking dataset, to tackle the lack of well-annotated datasets in this field^[1]. FEMOT covers diverse real-world scenarios and 14 challenging attributes, providing a reliable platform for evaluating RGB-event multi-object tracking methods. Additionally, they proposed FEMOTR, a multimodal tracking framework that decouples RGB and event features and fuses them in the frequency domain. Extensive experiments on FEMOT and DSEC-MOT datasets demonstrated the effectiveness of FEMOTR. Meanwhile, another research team proposed a new algorithm for efficient online 3D multi-camera multi-object tracking and pose estimation, requiring only 2D bounding box and pose detections^[2]. This algorithm is an efficient implementation of a Bayes-optimal multi-object tracking filter and eliminates the need for costly 3D training data or computationally expensive deep learning models. It is significantly faster than state-of-the-art methods without compromising accuracy and uses only publicly available pre-trained 2D detection models.