Training a Disaster Victim Detection Network for UAV Search and Rescue Using Harmonious Composite Images

Abstract

Human detection in images using deep learning has been a popular research topic in recent years and has achieved remarkable performance. Training a human detection network is useful for first responders to search for trapped victims in debris after a disaster. In this paper, we focus on the detection of such victims using deep learning, and we find that state-of-the-art detection models pre-trained on the well-known COCO dataset fail to detect victims. This is because all the people in the training set are shown in photos of daily life or sports activities, while people in the debris after a disaster usually only have parts of their bodies exposed. In addition, because of the dust, the colors of their clothes or body parts are similar to those of the surrounding debris. Compared with collecting images of common objects, images of disaster victims are extremely difficult to obtain for training. Therefore, we propose a framework to generate harmonious composite images for training. We first paste body parts onto a debris background to generate composite victim images and then use a deep harmonization network to make the composite images look more harmonious. We select YOLOv5l as the most suitable model, and experiments show that using composite images for training improves the AP (average precision) by 19.4% (15.3%→34.7%). Furthermore, using the harmonious images is of great benefit to training a better victim detector, and the AP is further improved by 10.2% (34.7%→44.9%). This research is part of the EU project INGENIOUS. Our composite images and code are publicly available on our website.

Systematically Improving the Efficiency of Grid-Based Coverage Path Planning Methodologies in Real-World UAVs’ Operations

“…Finally, out of the average results a radar chart was generated to provide a qualitative presentation of the modes at a glance. For the five non-normalized metrics, each mode is ranked in a scale of [1,5], where 1 corresponds to the worst value among them (not necessarily the smallest) and 5 to the best one (not necessarily the largest). This chart helps to better understand where each coverage mode outmatches the others and what is sacrificed to achieve that.…”

“…Recently, enterprise domains have been exploiting UAVs to collect specific types of data and perform some specialized tasks. Currently, some of the most common applications of UAVs' remote sensing are (i) precision agriculture [1][2][3], (ii) search and rescue [4,5], (iii) infrastructure inspection [6,7] and (iv) surveillance operations [8,9]. To make such operations more efficient in terms of resources' usage and quality of gathered data, the UAV missions are usually automated with a mission planner (such as Ardupilot Mission Planner (http://ardupilot.org/planner, accessed on 14 June 2023), Pix4DCapture (https://www.pix4d.com/product/pix4dcapture, accessed on 14 June 2023), Drone Harmony (https://droneharmony.com, accessed on 14 June 2023) and DJI Flight Planner (https://www.djiflightplanner.com, accessed on 14 June 2023) generating paths for the vehicle(s) participating in the mission, indicating the route to be followed and the specific points where data should be collected.…”

AI Enabled Robot for Data Collection in Unreachable and Extreme Environments: A Review

“…Ning Zhang, Francesco Nex, George Vosselman and Norman Kerle, [1] introduces human detection of images using deep learning has been a popular research topic in recent years and has achieved remarkable performance. Training a human detection network is useful for first responders to search for trapped victims in debris after a disaster.…”

Designing UAV Swarm Experiments: A Simulator Selection and Experiment Design Process

“…The figure shows the Crazieflie [78] agents in a swarm converging on the person. Target search and detection applications such as this open a range of possibilities in disaster management and emergency response [5,85,86].…”