Dronexscan: Detection and Identification of Vehicles in Aerial Images Captured by Unmanned Aerial Vehicles(UAVs)

This research study proposes Drone-TOOD, the low-power consumption task-aligned object detection algorithm for vehicle detection that was designed from the images captured by UAVs. In the aim of bringing low-power, real-time use into this methodfrom the ESP32-CAM image-processing module, it brings the principles of TOOD into it to balance precision and velocity while detecting vehicles through aerial views and challenges such as variation in altitude, lighting conditions, and orientations of the vehicles. In this regard, the low computational capabilities of ESP32-CAM are supplemented by optimized deep learning methods offering high performance in effective detection with minimal power consumption. Drone-TOOD has performed outstandingly well within the domain of UA V-based vehicle detection and hence asserts suitability for surveillance, traffic monitoring, and search-and-rescue applications.

I. Introduction

Recently, the application of UAVs has become widespread in many fields [1], such as response, traffic monitoring, and surveillance work. Recent advancements in computer vision point to an important need for aerial image object recognition techniques that are more effective and reliable. The difficulty of vehicle identification is primarily because of the variations in altitude, illumination conditions, and orientations of vehicles.