Every year SAE holds a Aerothon in which the participants are given certain tasks that need to be completed onboard a UAV/UAS.
The task for Aerothon 2022 was to "Develop an autonomous aerial system that can detect a target on a field, among dummy targets, and is able to drop a payload on the center of the target."
- Detect the target in the open field
- Find the center of the target and calculate the GPS coordinate for the same
- Make the drone fixate on the GPS coodinate
- Repeat the first 3 steps, to reduce error
- DROP!!
The entire system was deployed on a Raspberry Pi.
A default flight path is pre-set, during which when a taget is detected, certain commmands are run. The Models folder contains a custom implementation of UNET proposed by Ronneberger et al. using Tensorflow. The model was trained over a custom dataset of the target which was shot through a drone. From the output of the model, we first find the last and first pixel of the mask which allows us to find the center of the target(i.e. square) using helper.py. We then calcuate the GPS coordinate. GPS lock is set and the process is continued for n number of iterations to reduce the random error. Then the payload in dropped to the center of the target.🤞