Through crowd density estimation, an estimate for the number of individuals in an image can be obtained. State-of-the-art methods yield precise estimates by utilizing deep learning architectures exclusively. Aerial imagery is of special interest in this context because counts for entire areals can be obtained, in order to ensure for the safety of visitors at mass events. However, this field has been thus far researched only little, hence this work aims at doing so. Furthermore, a severe drawback of existing crowd counting methods is that estimates for image sequences are noisy, due to temporal dependencies not being exploited. Therefore, methods to overcome this issue are presented in this work, which yield accurate estimates that are also smooth over time.
Master of Science in Computer Science. Full-time nerd. Faible for bad puns.