Anticipating Human Activities from Surveillance Videos - A Survey

An important aspect of human vision is anticipation (prediction), which humans use extensively in day to day activities through interactions. Predicting which activities will be happening next, in unconstrained surveillance videos will help in monitoring and pin pointing the malicious activity prone areas. Anticipating activity is a probabilistic process - from the videos containing frames of onset of activity; the goal is to predict all possible futures. Prediction also helps in precise classification of activities in learning environments. Major work includes: (i) Action Recognition. (ii) Activities Classification. (iii) Developing the learning algorithms. And (iv) Predictions. Activities are modeled by Integral Histogram of Spatio Temporal Features and how the feature distribution change over time is captured. This paper focuses on the methods used in Person identification from video frames, such as Histogram of Oriented Gradients and Motion Boundary Histograms and how the feature extraction helps in action recognition and classification. The Bag of Words approach is studied to gain knowledge on prediction algorithms.