A real-time person-detection project based on darkflow (here)running on Nvidia Jetson TX2 Develope Kits.
The project aims at using a portable vehicle to detect and track people automatically. The scene will be displayed remotely.
This project began at Sep, 2018, and completed at Dec, 2018.
Python3, tensorflow 1.0, numpy, opencv 3, darkflow.
-
Getting the dependencies ready.
-
Following the guide here to install darkflow. You can also learn how to use darkflow in this web.
-
Run the FINAL.py if you have both vehicles and TX2 board, you need also change the "app.run()" line to your IP Address.
You can have ways dealing with the model weights.
-
Using the pre-trained YOLO weight files. Several kinds of YOLO can be choosed. In this project, as we only care about the category of person, the smallest net, tiny-yolo-voc, is recomended. It's weight file is in /bin. You can find other weight file here or here, and put them in /bin.
-
Train the model yourself. In this project, we are going to train our model.
That's all. Thank you for reading.