This project tackles the concept of recognizing and classifying heat patterns in order to detect humans. The heat data is collected from a thermal sensor attached to the ceiling of a room. It measures any object which emits heat. For example, the object could be an oven, human being or any other heat-emitting object. In order to detect humans, a model shall be created to distinguish heat patterns. The hypothesis of this research is that humans can be distinguished from other heat emitting objects using a thermal sensor. As such, the project is an object detection task.
Presentation: 13.12.2021
Final paper: 07.01.2021 23:59
https://www.overleaf.com/project/617665ffbc94254cf8ec4a41
https://www.overleaf.com/project/6188f27ebcae56e529194e55
https://docs.google.com/presentation/d/1TLgBtyyDolK1EsZuQ53GfDV2xGOMiX9MA92Uc6CJ1BQ/edit?usp=sharing
All raw data is now in the repo. It is just a few Megabytes.
We consented to use data_processing/step1-data-understanding.ipynb.
The sensors are treated independently, i.e., pictures with (almost) equal time stamps are not related to each other.
For manual annotation we show preceding and subsequent pictures to make the judgement easier.
We can derive binary classification (Human/Non-human) easily from the finer discrimination.