Human behavior recognition method based on conditional neural field

In this paper, a human action behavior method is proposed to identify the behavior of a single person on a public data set. After comparing different kinds of feature extraction algorithms, a robust adaptive visual background extraction algorithm is utilized to extract the algorithm feature. Then the centroid is used to intercept the target region and converted into a one-dimensional vector. Finally, we take advantage of feature vector for experiment training and testing. Comparing experimental result with that results of latent-dynamic conditional neural field model and support vector machine. The experimental result show that the conditional neural field model has higher recognition rate and better stability.