PLI stands for "Projet Libre Innovant" (Innovative free codecamp). It's part of our school program.

We want to analyze video streams for security interest. We want to detect dangerous behavior like the following :

• Fightings
• Fires
• People crossing the road when the light's red Oo
• Weapons
• People that have weapons but no uniforms

Our ultimate goal would be to detect habits of the people that are often captured by our CCTV cameras and to track them from one camera to another

We can use Machine Learning frameworks to perform object detection and then process the information.

Advantage : It's fast, well done and we don't have to deal with the difficult part (Mathematics, algorithmes ...). It's already optimized.

Inconvenient : We can do the project without understanding what the code (the framework) really does. Then we'll just learn how to use a framework and won't understand deep learning and neural networks as we want to.

We can read scientific papers about object detection and tracking and then implements algorithmes that researchers have developped.

Advantage : We'll be able to fully understand how object detection and tracking works. We'll learn how to implement algorithmes from scientific papers (that's a huge thing).

Inconvenient : It requires a lot of time and theorical knowledge.

We'll realize a POC (Proof Of Concept) with ML frameworks and then take the time to learn more about object detection algorithmes and try to implement them from scratch. Since we want to track and recognize people, we need to store a lot of data, especially images. So we need to choose a DBMS that meets our requirements.

DB : ext4 / btrFS File System to store images / video and neo4j for metadata