Thermal image recognition

The project was to design a tool, which would analyse data display from an IR sensor so that it can subsequently detect the presence of people in a room. Several cameras located in the area take thermal images and use them to determine how many people are there in the room and what are they doing. The activity of the persons is defined by few main states such as for example: seats, stands, moves.

Team members experience
Number of photos prepared during test set creation
0
After the learning phase the algorithm is capable of identifying the following states
STATE1

Algorithm indicates whether a person

alt
alt
alt
alt
sits
alt
stands
alt
moves
STATE2

Algorithm indicates whether a person

alt
alt
alt
alt
talks to someone else
alt
alt
talks on the phone

We found the project challenging in some areas. There was a need to choose adequate equipment components and develop software systems that could be capable of controlling the devices.

The final task was to design an artificial intelligence model, that would analyse particular thermal images. At this stage the data security and processing efficiency was crucial.

A prototype was made with the utilization of a Lepton FLIR sensor, which working with Raspberry PI, was functioning as a platform for testing and gathering data. Specific images gathered this way, were afterwards processed manually with employment of a purposefully designed software and computer analysed.

Each image out of 52.000 images was described- the number of people and their state was marked( stands, sits, moves) The dataset created in this way maximised the artificial intelligence learning process. That is the reason why the algorithm during the analysis is capable of identifying the number of people, even on new, not previously seen images.

As a result a new tool emerged, which on the basis of an image from thermal camera analyses movement of people in a room. It can also indicate, which parts of the space are being used more often, and where people prefer not to spend time. All of this takes place with a minimal invasion of privacy.

Find out how your company can benefit from Machine Learning