Traffic congestion estimation using HMM models without vehicle tracking

摘要: We propose an unsupervised, low-latency traffic congestion estimation algorithm that operates on the MPEG video data. extract features directly in compressed domain, and employ Gaussian Mixture Hidden Markov Models (GM-HMM) to detect condition. First, we construct a multi-dimensional feature vector from parsed DCT coefficients motion vectors. Then, train set of left-to-right HMM chains corresponding five patterns (empty, open flow, mild congestion, heavy stopped), use Maximum Likelihood (ML) criterion determine state outputs separate chains. calculate confidence score assess reliability detection results. The proposed method is computationally efficient modular. Our tests prove invariant different illumination conditions, e.g., sunny, cloudy, dark. Furthermore, do not need impose models for camera setups, thus significantly reduce system initialization workload improve its adaptability. Experimental results show precision rate presented very high- around 95%.