Large-scale event detection using semi-hidden Markov models

摘要: We present a new approach to recognizing events in videos. first detect and track moving objects the scene. Based on shape motion properties of these objects, we infer probabilities primitive frame-by-frame by using Bayesian networks. Composite events, consisting multiple over extended periods time are analyzed hidden, semi-Markov finite state model. This results more reliable event segmentation compared use standard HMMs noisy video sequences at cost some increase computational complexity. describe our reducing this demonstrate effectiveness algorithm both real-world perturbed data.