UNIMODAL: UAV-Aided Infrared Imaging Based Object Detection and Localization for Search and Disaster Recovery

Abstract

We propose a 5G ultra-capacity-aided, UAV-based, live streaming object detection and localization platform named ‘UNIMODAL’ (UAV aided iNfrared IMaging based Object Detection And Localization). We can not only live stream disaster and recovery scenes, but can also detect and localize humans or objects. In addition to using color images or video, it can detect and localize from infrared images and video with remarkable accuracy. We have trained various versions of YOLO including YOLOV3, YOLOV4 and the latest state-of-the art YOLOV7-official [1], and have achieved overall 95.62% mean average precision (MAP) using our object detection and localization model trained from YOLOV4. A detailed comparison between recent versions of YOLO has been performed; also the initial results using YOLOV7-official have been presented. The novel concept, detailed implementation, and preliminary results have been demonstrated in this paper.