Real-Time Survivor Detection in UAV Thermal Imagery Based on Deep Learning

Wide-Area and Real-Time Object Search System of UAV

“…Zhan W. et al [28] improved the YOLOv5 object detection algorithm from four aspects in order to achieve real-time detection of small objects, as follows: by redesigning the anchor size, adding attention module to the backbone, using CIOU loss function, and adding the P2 feature level [29] proposes ShuffleDet based on ShuffleNet [30], and a modified variant of SSD [19] to realize real-time vehicle detection by UAV. While improving the algorithm, there is also a part of research that describes the hardware composition and the workflow of the real-time object search system of UAV [31], which describes a real-time survivor detection system with a pruned object detection algorithm in a UAV, proposed in order to reduce the loss of lives caused by natural disasters. In [32], Chen, L. et al collected video stream and executed real-time object detection by carrying a camera, then controlling the UAV to perform corresponding actions.…”

“…Most of the UAV real-time object search systems that are involved in previous research [31,32,[36][37][38] are more suitable for low-altitude and short-endurance flight tasks, and the equipped equipment is often a small-resolution camera for taking images or videos; there is little research on real-time systems when the UAV flies at mid-to-high altitude. While many UAVs can be used in parallel to achieve wide-area coverage, this requires the addition of multiple UAVs and ground crews, resulting in a significant increase in the operating costs.…”

“…With the help of the parallel computing capabilities of the edge device, the system adopts the idea of parallel modular design to improve the system's performance. Based on this design, the system can also flexibly adapt different types of cameras and objection detection algorithms according to the task requirements, thereby increasing the versatility and Most of the UAV real-time object search systems that are involved in previous research [31,32,[36][37][38] are more suitable for low-altitude and short-endurance flight tasks, and the equipped equipment is often a small-resolution camera for taking images or videos; there is little research on real-time systems when the UAV flies at mid-to-high altitude. While many UAVs can be used in parallel to achieve wide-area coverage, this requires the addition of multiple UAVs and ground crews, resulting in a significant increase in the operating costs.…”

“…Otherwise, the object positioning will significantly deviate from the actual location. In many studies, such as [31,38], there are a lack of descriptions for object positioning, because the object occupies a large proportion of the image frame at a limited Remote Sens. 2022, 14, 1234 4 of 23 altitude, in which it can be approximately considered that the object location is equivalent to the location of the UAV.…”

“…Otherwise, the object positioning will significantly deviate from the actual location. In many studies, such as [31,38], there is a lack of descriptions for object positioning because the object occupies a large proportion of the image frame at a limited altitude, which can be approximately considered that the object location is equivalent to the lo-cation of the UAV [37] mentioned the study of velocity estimation of the tracked ob-ject, but also no precise coordinate localization. This paper will also introduce the detailed calculation method to perfect the whole system.…”