Multi-Camera Trajectory Forecasting: Pedestrian Trajectory Prediction in a Network of Cameras

We introduce the task of multi-camera trajectory forecasting (MCTF), where\nthe future trajectory of an object is predicted in a network of cameras. Prior\nworks consider forecasting trajectories in a single camera view. Our work is\nthe first to consider the challenging scenario of forecasting across multiple\nnon-overlapping camera views. This has wide applicability in tasks such as\nre-identification and multi-target multi-camera tracking. To facilitate\nresearch in this new area, we release the Warwick-NTU Multi-camera Forecasting\nDatabase (WNMF), a unique dataset of multi-camera pedestrian trajectories from\na network of 15 synchronized cameras. To accurately label this large dataset\n(600 hours of video footage), we also develop a semi-automated annotation\nmethod. An effective MCTF model should proactively anticipate where and when a\nperson will re-appear in the camera network. In this paper, we consider the\ntask of predicting the next camera a pedestrian will re-appear after leaving\nthe view of another camera, and present several baseline approaches for this.\nThe labeled database is available online:\nhttps://github.com/olly-styles/Multi-Camera-Trajectory-Forecasting.\n