Counting the number of people who enter a shop every day by using the video streams from CCTV Cameras.
To have a reliable and pragmatic solution for this problem, we should consider a few main points:
- Hardware capacity: most algorithms that use a Deep Neural Network to solve this problem require a lot of resources. Limited processing power usually creates a trade-off between the precision and responsiveness of the algorithm. The most precise algorithm may run at one frame per second (1FPS), and if you require a faster response (e.g. 10 FPS), you must either choose a less precise algorithm or spend more to get better hardware.
- Counting everyone only once: When someone goes out of the shop and returns after a few seconds, you should count them as one person. This can be tricky in real-world scenarios. You should use feature extraction algorithms to assign a vector to each person. When a new person enters the frame, you should check to see if they are new or returning.
This repository consists of seven Jupyter Notebooks that walk the reader through different steps of solving this problem. It also includes some sample codes for each step.