AEOLUS – AI-driven Efficient Object Detection for Low-power Unmanned Surveillance

Project Title: AEOLUS – AI-driven Efficient Object Detection for Low-power Unmanned Surveillance

Programme: PHD IN INDUSTRY
Proposal Number: PHD IN INDUSTRY/1123/0145
Duration: 01/11/2024 – 30/10/2027 (36 months)

Funding Agencies:  The Project PHD IN INDUSTRY/1123/0145 is funded by the Republic of Cyprus, through the Research and Innovation Foundation.

Unmanned aerial vehicles (UAVs) are often limited by size, weight, and power constraints, leading to cheaper, lightweight equipment and subsequently lower imaging capabilities. Recent advances in deep learning (DL) detection models have made it possible to achieve real-time or near-real-time object detection. However, current state-of-the-art continues to favour larger models on cloud-based solutions rather than edge-enabled and embedded systems to avoid meticulous deployment optimizations. This project aims to provide a balanced solution to the trade-off between performance and inference speed by developing lightweight yet powerful object detection models for UAVs to overcome the limitations of current imaging technology due to cost, operational, technological, and computational factors. Object detection can be employed to gain valuable insights into objects of interest and their spatiotemporal status in the input stream. In UAVs it can equip users with advanced imaging capabilities, leading to better decision-making in various fields. The proposed project will realize this objective by designing, developing and training DL models for UAVs. Optimization schemes for real-time or near-real-time performance on low-power devices will be intensively explored. In conclusion, the project’s expected results include edge-deployed object detection models for UAVs, delivering advanced image analytics and reducing computational as well as operational costs. The socio-economic impact of this project includes enhanced surveillance and monitoring capabilities, empowered search and rescue operations and better decision-making in smart city applications such as law enforcement, emergency response and monitoring traffic and pedestrian safety.