2nd Workshop on Benchmarking Trajectory Forecasting Models
October 16, 2021
Aims and Scope
Human motion prediction is an essential task for developing autonomous systems able to safely navigate public spaces or interact with humans. Navigating crowded scenarios requires mimicking human perception and extracting social conventions in order to autonomously move in these areas. Various types of sensors (e.g., LIDAR scanner, RGB camera, GPS) are employed to perceive the surrounding environment by self-driving cars or robots. Another important area where forecasting human dynamics is used is advanced video surveillance systems for detecting anomalous behaviors or promptly supporting human operators. In the past few years, several methods have been proposed to predict human motion yet they have partially addressed how to evaluate their results in case of multi-modal predictions and critical situations mainly because no proper benchmarks were tailored for efficiently tackle this problem.
Despite its potential real-world applications in several domains (e.g., computer vision, robotics, healthcare), this task has not received adequate attention as detection or recognition problems. In this workshop, we aim to discuss recent advancements in this field with researchers in computer vision, robotics and cognitive neuroscience areas to conceive autonomous systems able to proactively act in complex contexts involving humans and moving objects in a safely manner. We will also lay the foundations for future research, powering discussions on applications, data and performances.
Human trajectory and activity forecasting in urban scenes
Evaluation and benchmarking in motion prediction
Crowd motion analysis
Human dynamics modelling
Visual scene analysis
Path planning and optimization
Data fusion techniques