Planned intervention: On Wednesday June 26th 05:30 UTC Zenodo will be unavailable for 10-20 minutes to perform a storage cluster upgrade.
Published December 31, 2021 | Version 1.00
Project deliverable Open

Data-driven analytics applied on UAV imagery using deep learning

Description

This deliverable presents the overall development status of the deep learning analytics applied on UAV imagery (both visual and thermal) on M18 of the project’s lifetime. Within the duration of T3.3, several DL pipelines were designed, implemented, and tested on the task of object detection with humans as the main object of interest. Different variants of the pipelines were investigated including various powerful State-of-the-Art object detection network including Yolov4, scaled-Yolov4, Yolov5, Detectron2 and FasterRCNN. In addition, a novel hybrid inference mechanism was proposed, developed, and tested to cope with the identified challenges especially with respect to the effect of the UAV flight altitude. The proposed inference mechanism combines the output of altitude-dependent local deep learning increasing the generalisation capabilities of the OD system. An extensive experiment set-up was designed to identify the best performing deep learning networks and demonstrate the detection performance of the proposed object detection pipeline utilizing both spectral and thermal information.

Files

SnR_D3.5 - Data-driven analytics applied on UAV imagery using deep learning_v1.00.pdf

Additional details

Funding

Search and Rescue – Search and Rescue: Emerging technologies for the Early location of Entrapped victims under Collapsed Structures and Advanced Wearables for risk assessment and First Responders Safety in SAR operations 882897
European Commission