Visual analysis of urban road traffic

The authors discuss the problem of understanding urban traffic motion flow in noisy environments. We propose pixel based modeling of motion directions with the help of Gaussian mixture models (GMMs) and hidden Markov models (HMMs). This way there is no need for object tracking and the proposed modification in the original hidden Markov model gives us a more usable tool for the analysis of hundreds of samples at a time. The proposed methods can be used for visualization and anomaly detection purposes.