University practical work to implement a basic people tracking using predictive Kalman Filter.
Is has been done for the Course Credits VA53 - Probabilistic and stochastic models for computer vision at Université de Technologie de Belfort Montbéliard (FR).
The aim was to try different models (Position, Speed, Acceleration) at different video input framerate.
As the body detection was not the main goal of this work, it has been done roughly and is not fine-tuned for every video input.
The report written for evaluation (in french) as well as the measurements sheet are available under /docs.
The images generated for illustration purpose are under /img, where Green line is the prediction and Red line the detection.
Note : the input data used in this project has been removed for privacy purpose.
Just run kalman.py with Python 3.
You can input file and switch the Kalman model by editing inside the source code.