This repository extends a lab from the GPA659 course at ÉTS Montréal. It implements motion detection using a Gaussian background model (per-pixel mean and standard deviation). After preprocessing (masking noisy regions and applying an averaging filter), the method computes the normalized squared deviation pixel per pixel, aggregates it into a global motion metric, and compares it to a threshold. The background model is trained on 80% of the dataset (first 960 frames) and then evaluated on the full sequence.
Clone the repository:
git clone https://github.com/nellbru/motion-detection-matlab.git
cd motion-detection-matlabOpen MATLAB and run model_parameters.m to build the background model (trained on frames 1–960).
Then run test_motion_detection.m to evaluate the model on the whole dataset and display the motion detection.
The visualization shows:
- source image
- motion per pixel (log scale) with a colorbar
- motion (log scale) within the whole image over time
The script also reports summary counts (realMotionCount, modelMotionCount, diff) and overall accuracy, which is 95.5%. The remaining 4.5% error is inherent to the background modeling method used, new but stationary objects in the scene (for example a car that stops) are treated as motion because they differ from the learned background.
motion-detection-matlab/
├─ data/ # Input frames
├─ docs/ # Documentation assets
├─ labels.mat # Ground truth labels
├─ model.mat # Saved background model (generated)
├─ model_parameters.m # Script to generate the model
├─ motion_detection.m # Detection function
├─ test_motion_detection.m # Script to run the full pipeline
├─ .gitignore # Ignore rules for Git
├─ LICENSE # Project license
└─ README.md # Project documentation
- MATLAB (R2018b or newer)
This project is licensed under the terms of the LICENSE file.