M600 Drone Custom Thermal Detection Payload With Gimbal Using Tensorflow Lite
This Project is currently not in its full state. It currently is a proof of concept that I wish to take further with updates over time.
How to set up on Raspberry Pi for object detection (Note: Only tested on Pi 5 with Bookworm OS as of 4/1/2024)
Installation on a Fresh Install of Bookworm, please set up an internet connection beforehand
Type these commands into Command Terminal
Set up a VM if you want, I did not as this is the only purpose this Pi will be used for currently
sudo apt-get update
sudo apt-get upgrade
sudo rm /usr/lib/python3.11/EXTERNALLY-MANAGED (stops externally managed error)
sudo pip3 install opencv-python
sudo pip3 install mediapipe
That is it for setup.
Please look at 3 Files:
FLIR SETUP with Assistant 2 and FLIR UAS
Mounting the RPI, FLIR, and Buck Converter (WIP)
Final Connections and DJI Go setup (WIP)
Once you have gone through the above setups. Open the repository on your Pi.
Open FLIR-Detection.py
in Thonny or your Python runner of choice.
Run the program with the Flir Connected to the Pi. This will start the detection. For testing a USB webcam may work, but the model is not tuned for that.
Open Gimbal.py
to start the Servo's balancing in Thonny or your Python runner of choice.
Make sure the MPU6050 is connected and the Servo before running. Yous should see the Gimbal lock to a certain pitch.
If you wish to do this, please take a look at the TRAIN YOUR OWN MODELS file if you want to capture/performance tune your own model set.
Note: The training method I use is using Google's Colab as I didn't want to set it up on my own machine.