Person Detection in Aerial Images

Abstract

With advancement in technology, a lot has changed. Starting from the daily life events like farming to mega events like launching satellites. Industrialization has changed our world altogether. For most of the times, it has changed us for good. Today, the average life expectancy age is 2 times more than that of what it was a few centuries ago. Technology has played its part in it and to some large extent.

Previously, once a flood inundated an area, or any other natural disaster occurred, it left only after inflicting hundreds of causalities. We, the team Person Detection form in Aerial Images, want to change that. The primary problem in such cases is detecting where the affected person is. We are developing an android app and a web app that will help concerned agencies detect if there is any person at the particular location through drones. This particular location can be a fire-affected-area, flood-affected-region, an enemy’s military camp, or a place that cannot be accessed directly. Once the authorities have detected the people on ground, the subsequent operations can be carried out accordingly.

Moreover, aerial monitoring serves as an important and essential ingredient for many applications such as surveillance, reconnaissance, and disaster recovery and rescue procedures. Prior to scene analysis and interpretation, a common requirement for all such applications is the need of robust algorithms able to automatically detect relevant objects within the field-of-view of the optical sensor. Such automatic detection not only reduces the image analysis work load but also paves way towards developing intelligent systems based on the state-of-the-art machine (deep) learning algorithms having immense application potentials.

In this project, we aim to detect persons/people using state-of-the-art deep learning techniques from data acquired from publically available datasets and/or a low-cost low-

altitude unmanned aerial vehicle or commonly referred to as drone such as DJI Phantom pro. The detection of person(s) yields advantages in automatic action/anomaly detection, devising escape plans in crowds etc. The work done in this project will be a subpart of problems to be solved within the scope of newly established deep learning lab under the umbrella of National Center of Artificial Intelligence (NCAI).