Multi-Person Detection and Segmentation with State-of-the-Art Deep Learning Methods (Bachelor Project, Finished)


Claire Walzer


Understanding human behavior in social interactions by machines is one of the still not fully covered fields in computer vision and is essential to use concepts of social interaction as a source for searching in large collections of multimedia objects. One of the key elements in these social behaviours which crucially impacts the way a machine can respond to certain approaches is gesture recognition. In many scenarios, the streaming input to the machine does not include only one person, but rather multiple persons, occasionally occluding each other. The challenge in such scenarios is to be able to extract, detect and segment each person with high accuracy to further be able to analyze their gestures via the gesture recognition methods.

This project specifically focuses on implementing and using the existing state of the art algorithms to detect and segment people, and compare these algorithms for gesture recognition task.

Start / End Dates

2020/02/17 - 2020/07/06


Research Topics