People detection (and tracking) with sparse 3D data
Robust people tracking is an important component in robotic systems that are to work alongside humans. The goal of this project is to train a classifier to detect persons from sparse 3D data, such as the output from a Velodyne VLP-16 scanner. The project will consist of training a classifier on segmented data and study the limits at which it can reliably detect people (range from sensor, amount of occlusion). If time permits, the project may also include using the people classifier for tracking and/or training it on data from non-upright people (sitting, crouching, etc).
Contact: Tomasz Kucner