A Data Annotation Process for Human Activity Recognition in Public Places

Behavior analysis of individuals in crowds or groups of people in public
places through surveillance cameras gains importance for several different actors. Automatically detecting and understanding pedestrians in real-world uncooperative scenarios is very challenging. Common issues such as limited annotated data, unreliable data and annotation quality, and appropriate use of this data for supervised learning often originate in steps preceding the modeling of specialized neural network architectures. In this report, the necessity and requirements for designing a reliable data annotation process are presented. Some precise ideas for automation through neural networks are discussed in a conceptual manner.