| Summary: | 碩士 === 國立臺灣海洋大學 === 資訊工程學系 === 106 === The problem of detecting human objects in a crowd in photography lies in how to accurately capture objects’ features and how to continue tracking their trajectories among a crowd. The human form detection accuracy is affected by different environments and movements such as shadows, light, and human postures. Trajectory tracking with human objects is interrupted by shadowing and overlapping, which increase the complexity of trajectory tracking. Aerial images taken by continuously moving aerial photography drones also have a certain degree of influence on tracking the human form. This paper attempts to overcome these problems by proposing a method of accurately tracking the human form in ongoing aerial photography and analyzing crowd density to detect crowd hotspots.
The advent of deep learning expands new horizons in image processing; it solves many difficult problems in various fields and drives the development of relevant industries. This paper also employs deep learning techniques to deal with human form detection problems. Hopefully, these paper results can find further applications in aerial photography.
|
|---|
