Inexpensive Search & Rescue UAV with Mission-Planning and Machine Learning Human Detection

Dennis Pavlyuk, Foluso Odeyale, Ryan Ellis, and Dr. Ning Yu, Department of Computer Science, State University of New York Brockport, 350 New Campus Dr, Brockport, NY 14420

There are many unmanned aerial vehicles (UAV's) available for search and rescue missions in the consumer market. The UAV industry is oversaturated with gyro-stabilized multicopters that support cameras attached to gimbals, have an artificial-intelligence-assisted flight, and have digital video transmission from impressive distances. The problem is that the majority of them are prohibitively expensive to own for small governments and municipalities.  While these drones allow for high-definition video, the attachment of additional sensors, and stable flight, their short flight-times and exorbitant prices make their utility limited. We seek to show a solution to this problem by proving the efficacy of an inexpensive Fixed-Wing UAV. 

We propose to make a relatively-inexpensive fixed-wing unmanned aerial vehicle that uses a model-based object-detection algorithm (tinyYOLOv3) to interpret images in real-time to identify humans on the ground. There has been a previous attempt to create a similar fixed-wing search and rescue UAV by Johnatehn Mendenhall, but it failed due to a.) pilot error, b.) excess weight, and c.) unoptimized design of the aircraft.  We have addressed problem b.) by replacing the UAV's companion computer, an NVIDIA Jetson Nano, with a Raspberry Pi Zero W and Intel Movidius NCS, and also by replacing the UAV's companion computer's external battery pack with a connection to the power-distribution board of the flight controller. This is a total weight reduction of 370 grams. We have addressed problem (c) by prototyping with consumer-grade aircraft models used specifically by RC FPV hobbyists such as the ZOHD Nano Talon and Flite Test Blunt-Nose Versa Wing. There has been reported success with creating a search and rescue UAV using the ZOHD Nano Talon by MKME Labs. Still, they use Access Point Beacon Frames identified with an ESP8266 chip to locate individual mobile devices rather than an image object-detection approach of a bottom-facing camera.

Additional Abstract Information

Presenters: Dennis Pavlyuk, Ryan Ellis

Institution: State University of New York- Brockport

Type: Poster

Subject: Computer Science

Status: Approved

Time and Location

Session: Poster 5
Date/Time: Tue 12:30pm-1:30pm
Session Number: 4007