| Summary: | This final-year project explores the feasibility and effectiveness of the Point-Query Quadtree (PET) crowd estimation method on a localised (Singaporean) dataset. As borders open and crowd restrictions ease, there is need for such technologies to prevent an onset of crowd crush or stampede situations, which although may be few and far between, but can have devastating long-term consequences to many people. Two extensions to the PET method incorporating depth estimation using the DepthAnything framework will be assessed and analysed for its efficacy and improvements. In the modifications done, we show that a 33% decrease in mean absolute error is possible, honing the scalability and effectiveness of PET and with modifications.
|
|---|