Ampel2Go - An algorithm to count the amount of bidirectional movements
We have built a python library that is capable of detecting moving objects and counting the direction in real-time using minimal computing power (i.e. it runs on a RaspberryPi 4b)
After our extensive market research we have concluded this to be one of the best performing open source people counter using low cost edge-computers.
Content:
- 101_ai_module: This is the main and core part of our work containting the movement detection and counting algorithms
- 102_user_display_and_settings_module: A django application that reads relevant values (counts, etc. ) from the central database, displays it, and enables changes via the settings screen.
- 103_remote_control_module: This is a second option to changes settings accomanying the settings module using a infrared-remote-control
- 201_sample_videos: Contains sample videos to be referenced in 101_ai_module
- 204_logs: Folder containing logs
How to get started with the detection algorithm using pretrained models
- Clone repo and cd into the folder
- Start the pipenv ("pipenv shell") (the pipfile is located in the parent directory)
- Run this command: "python3 101_people_counter/main_people_counter.py --input 201_sample_videos/olash3h043_testtype-norma-height-240-up-000-down-001.avi"
- Three windows will pop up displaying the following:
How to get started with the labeling application
You can train the model to adapt to special situations. In our sample-videos, the model is trained to ignore the moving door that enters the picture from below. Entry point is the file: main_generate_training_data.py, which starts the labeling procedure.
How to get started with the user display and settings module
- Clone repo and cd into the folder
- Start the pipenv ("pipenv shell") (the pipfile is located in the parent directory)
- Run this command: "python3 102_user_display_and_settings_module/manage.py runserver"
- In your browser, go to "0.0.0.0:8000" to see the frontend of the application:
