Real time object detection for aerial search and rescue missions for missing persons
Zsolt Domozi, Dániel Stojcsics, Abdallah Benhamida, Miklós Kozlovszky, András Molnár
Abstract
This paper introduces a solution to stand-alone system based, real-time object-detection, can efficiently facilitate the search for missing persons with an unmanned aerial vehicle. The challenge is the real-time implementation of the systems and training the given deep neural network for the desired task. The paper describes the methods and procedures currently in use, as well as the possible tools. Subsequently, the autonomous aircraft system, which carries a real-time detection system, is introduced. In the section about real-time detection, we will introduce the TensorFlow lite-based application, building on SSD topology, in detail, which was implemented on mobile phones. We will also introduce the dataset used for training, testing and the results achieved. In summary, the recall achieved is 65.4% and precision is 96.4%, besides the fact that the android-based application, using the phone’s camera, performs image analysis at a rate of 11 to 17 FPS in real-time, while continuously providing