Analysis of Human Walking Trajectories for Surveillance
This chapter provides an application in which the actions of humans are modeled for analysis. It explores the artificial intelligence and statistical analysis techniques towards real-time observation of people, leading to the classification of their walking patterns. Real-timeliness is especially important for the application of surveillance, in which crimes and abnormal activities can only be prevented and stopped if the system is equipped with the capability to understand human actions at the moment when they are performed. The criteria embedded in the trained support vector machines will be applied to the classification of new observed data in real-time. Dynamic time warping has been applied in a number of application areas which involve data comparison, including DNA sequencing, handwritten text classification, and fingerprint classification. For global human walking trajectory classification, the performance of the Hidden Markov Models-based similarity measure is not as good as the best longest common subsequences-based approach.