On Flight Real Time Image Processing by Drone Equipped with Raspberry Pi4

Abstract

This paper presents an innovative solution for image processing, especially object and person detection, using a drone. Compared to previous implementations that relied on a good network connection for good functionality, in my case the processing/detection will be done locally using a Raspberry Pi4 module, mounted on the drone. The first part of our research consisted of building a lightweight drone with some of the best racing components, chosen carefully for our specific needs. The second part consisted of implementing a computationally-low-cost detection algorithm on the Raspberry Pi, using convolutional neural networks and Tensorflow Lite. The chosen algorithm that was used was SSD - Single Shot Detection, capable of detecting multiple objects in one image during one iteration of the algorithm. I also designed an enclosure to securely place the drone battery, Raspberry Pi Battery and Raspberry Pi module on the drone, which was 3D printed.