Parallelizing abnormal event detection in crowded scenes with GPU

Analyzing human activities in surveillance videos a challenging task due to the high volume of data that needs to be processed in a timely manner. This paper presents a GPU based Gaussian Mixture Model (GMM) to detect abnormal activities in crowded scenes. GMM is a fully unsupervised method that predicts abnormal crowd behaviors based on the processing of normal crowd behaviors. Specifically, we use crowd distribution and GMM to estimate the speed and to predict the behaviors of the crowd. The performance of the parallel GMM is evaluated from the aspects of computation efficiency and accuracy in terms of area under the curve.