A new two-stage topic model based framework for modeling traffic motion patterns
Parvin Ahmadi, Iman Gholampour, Mahmoud Tabandeh · 2017
Analyzing motion patterns in traffic videos can be exploited directly to generate high-level descriptions of the video content. The most recent and successful unsupervised methods for complex traffic scene analysis are based on topic models. In this paper, a new two-stage framework is proposed for traffic motion pattern extraction based on topic models. This framework forces the topic model to learn known meaningful motion patterns in traffic scenes. Latent Dirichlet Allocation (LDA) is employed as the topic model. Experimental results show that our proposed framework finds the motion patterns more efficiently and gives a meaningful representation for the video.