Workshop Goals: Dataset and Challenges

This workshop focuses on the unique perceptual problems related to autonomous navigation in indoor and outdoor human environments. These problems include human detection and tracking from 2D and/or 3D data, human posture detection and prediction, object detection, segmentation, trajectory forecasting and any other perceptual task that, when solved, provides valuable information to autonomous agents and robots that need to navigate safely in human crowded environments.

Last years the community has paid increasing attention to these perceptual problems and has provided annotated datasets to learn to solve them and evaluate them, yet restricted to the point of view of autonomous vehicles. Robots and other navigating agents, however, view the world from a different perspective, can go indoors, get closer to humans and need to perceive other types of information. In this workshop, we present the JackRabbot social navigation dataset , a novel annotated dataset with the signals from our mobile manipulator JackRabbot. For the workshop we launch the first set of visual challenges, to benchmark perceptual algorithms in this new type of data.

The dataset contains 67 minutes of ground truth annotated sensor data acquired from the JackRabbot mobile manipulator and includes more than 50 indoor and outdoor sequences in a university campus environment. Sensor data includes a stereo RGB 360° cylindrical video stream, 3D point clouds from two LiDAR sensors, audio and GPS positions.

In the workshop we will present the winners of the first challenge. We have also as invited speakers world renowned expers in perception for autonomous navigation. We aim to foster discussion between the attendants to find useful synergies and applications of the solutions of perceptual tasks for navigation in human environments.

Dataset

In this first phase, the annotations of the JackRobot Dataset include:

2D bounding box annotations around pedestrians, cyclist and vehicles in five RGB streams and in their cylindrical composition.

3D oriented cuboid annotations around pedestrians, cyclist, vehicles in two Velodyne-16 LiDAR streams

2D-3D associations between bounding boxes and cuboids

Time consistent trajectories (tracks) for all annotated objects in both 2D and 3D

Open Challenge

We launch in this workshop a first challenge on our annotated data in the following categories:

2D-3D person detection

2D-3D person tracking

The winner of each challenge will be given an opportunity to present their work as oral peresentation during the workshop.

Call for Papers

We invite researchers to submit their papers addressing topics related to autonomous (robot) navigation in human environments. Relevant topics include:

2D or 3D object detection and pose estimation, tracking and forecasting

2D or 3D pedestrian and vehicle detection, tracking and forecasting

2D or 3D human skeleton pose estimation, tracking and forecasting

2D or 3D human face detection

Human emotion recognition

Human walking behaviour analysis

2D and 3D semantic, instance or panoptic segmentation

Motion prediction and social models

Individual, group and social activity recognition

Abnormal activity recognition

Visual and social navigation in crowded scenes

Traversability estimation

Dataset proposals and bias analysis

New metrics and performance measure for different visual perception problems related to autonomous navigation

Submisison format should match the ICCV format, e.g. 8 pages, double column format. By submitting to this workshop, authors agree to the review process and understand that we will do our best to match papers to the best possible reviewers. The reviewing process is double blind. Submission to the challenge is independent from the paper submission, but we encourage paper authors to submit to one of the challenges.

Submissions can be made here. If you have any questions about submitting, please contact us here.