Extraction of Bicycle Crash Kinematics from Videos Using Machine Learning

Pose estimation example using OpenPose. Example bicycle crash

It is extremely difficult to obtain motion data of bicycle crashes for a number of reasons. But this kind of data would be very useful in understanding why falls and crashes occur during bicycling. Understanding the mechanisms are key to preventing injuries and death of bicyclists. It is a bit easier to find and collect video data of crashes and there are a number of new technologies that make it easier to extract kinematic motion and pose data from videos using machine learning (e.g. DeepLabCut, OpenPose, Anipose). The TU Delft Bicycle lab has a reasonably large set of training data of typical (non-crash) bicycling that can be used to build the video to kinematics predictive models. The goal of this project would be to collect a large set of bicycle crash videos, build a model that can extract the kinematics of both the bicycle and the rider, and show that accurate kinematics can be acquired.

How To Apply

If you would like to apply for this project, please send an approximately half page letter explaining your motivations and interest in the lab and project, CV or resume, a list of courses you've taken, the name of your MSc track, and any other relevant information to j.k.moore@tudelft.nl.