Detecção de Anomalias Em Vídeos Utilizando Dicionários Espaço-Temporais

This paper presents an implementation of the Spatio- Temporal Compositions (STC) method for the detection of video anomalies. The STC, as the Bag of Video words (BOV), uses a dictionary to eliminate to redundancies, but differs from it by taking into consideration the spatio-temporal composition of small volumes of videos. The STC also performs a modeling using a probabilistic approach, in which anomalous events are those with a low probability of occurrence. Some modifications incorporated to the STC include volume superposition level, spatio- temporal filtering stage and multi-scale computation of probability function. The algorithm used in this model gives good results in the identification of anomalies, without background subtraction, motion estimation or tracking. The system is accurate even with a little training and no prior knowledge of the type of event to be observed, being robust to light variations or cluttered environments, as illustrated by several examples.