Difference between revisions of "Projects:2019s1-111 Deep Learning-based Object Detection and Tracking of Moving Targets from a Drone"
(→Abstract) |
|||
| Line 1: | Line 1: | ||
| + | = Abstract = | ||
| + | The recent increase in image training datasets and GPU processing power have enabled the performance of machine and deep learning algorithms to become the benchmark standards for image classification and object detections. Many of these algorithms are now used in the autonomous car for real-time scene understanding and in the Google image search engine algorithm. | ||
| + | | ||
| + | Object Detection is the method of identifying real-world objects such as cars, trucks and people, within images and video. Typically an image is processed through an object detection algorithm which extracts features and a learning algorithm which identifies occurrences of an object class. | ||
| + | |||
| + | The aim of this project is to determine the performance of state-of-the-art deep learning object detection algorithms to detect and track a moving target from a fast moving drone. | ||
== Introduction == | == Introduction == | ||
| Line 23: | Line 29: | ||
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| | ||
Revision as of 14:33, 12 April 2019
Abstract
The recent increase in image training datasets and GPU processing power have enabled the performance of machine and deep learning algorithms to become the benchmark standards for image classification and object detections. Many of these algorithms are now used in the autonomous car for real-time scene understanding and in the Google image search engine algorithm. Object Detection is the method of identifying real-world objects such as cars, trucks and people, within images and video. Typically an image is processed through an object detection algorithm which extracts features and a learning algorithm which identifies occurrences of an object class.
The aim of this project is to determine the performance of state-of-the-art deep learning object detection algorithms to detect and track a moving target from a fast moving drone.
Introduction
The motivation behind this project is the need for moral weapons to become more prevalent in the defence force.
In July 2015 an open letter with signatories including Elon Musk, Steve Wozniak and the late Stephen Hawking asking for a ban on autonomous weapons was released. In summary, the letter asks for a ban on offensive autonomous weapons beyond meaningful human control. In addition to this, in 2013 the UN presented a report with recommendations for testing, production and deployment of LARs (Lethal Autonomous Robotics).
An example of why many world-leading academics feel so passionately about this issue is the Grdelica train bombing disaster which occurred in 1999 in Serbia. 2 missiles were fired by a NATO aircraft with the aim to bomb the bridge. The pilot claims to have not seen the passenger training and unfortunately hit the train killing up to 60 people. By implementing moral weapons, an autonomous loitering weapon (e.g., drone) it will be able to detect, identify and tracks key ground objects near and around a target to avoid events such as the Grdelica disaster.
Project Team
Student Researchers
David McDonough
Eleanor Harrihill
Project Supervisors
Dr Cheng-Chew Lim
Dr Carmine Pontecorvo (DSTG)
Dr Hiroshi Yokohama (DSTG)
