Melanoma is the most serious type of skin cancer in the world, and its prevalence has increased in recent years. Total Body Photography (TBP) is a well-known tool for early diagnosis of melanoma, which significantly improves the chance of its treatment's success. Current commercial TBP systems present several challenges, including their expense, bulkiness, and requirement for a large, dedicated location. To overcome these limitations, we propose a drone-based system, called "DermDrone," as an innovative concept for TBP. Firstly, a needs assessment procedure was conducted using off-the-shelf drones to recognize the user and device requirements. Afterward, the first MVP for DermDrone was proposed that offers an automated method for conducting TBP. The drone employs on-board real-time vision-based navigation using a customized floor map that allows it to follow a pre-defined trajectory around a patient and capture photos of all body parts. To improve the navigation method, this work proposes a human pose estimation (HPE)-based localization algorithm that provides positioning feedback with respect to the patient body without any external visual markers. Consequently, an RL -based method is implemented to enhance the accuracy of the proposed localization method by selecting the best viewpoint. Ultimately, an end-to-end RL-based navigation method is introduced in which DermDrone learns how to navigate around a patient using the result of HPE. The performances of both viewpoint selection and end-to-end navigation are shown with several experiments in simulation.
Copyright is held by the author(s).
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Supervisor or Senior Supervisor
Thesis advisor: Arzanpour, Siamak
Member of collection