A three-stage framework for smoky vehicle detection in traffic surveillance videos

Abstract The smoky vehicle exhaust pollutes the air and endangers human health. Existing methods detecting smoky vehicles in traffic surveillance videos are with high false alarm rates due to the diversity of smoke characteristics and continuous interferences of common vehicles. To solve this issue, this paper presents a three-stage framework for smoky vehicle detection. In the first stage, a Robust Pixel Based Adaptive Segmenter (R-PBAS) algorithm, which adapts to cameras shaking, is proposed to extract moving objects. The Cumulative Color Histogram (CCH) is adopted to extract smoke-colored blocks from moving objects. In the second stage, three groups of features, including Non-Redundant Robust Local Binary Pattern (NR-RLBP), Weighted Co-occurrence Histograms of Oriented Gradients (W-CoHOG), and Motion Boundary Histograms (MBH) are proposed to extract texture, gradient, and motion information from smoke-colored blocks, respectively. In the third stage, we fuse smoke blocks to obtain Region of Interest (ROI) and extract frequency domain features based on Discrete Wavelet Transform (DWT). To further improve robustness, the Auto-Regressive and Moving Average (ARMA) Model and Hidden Markov Model (HMM) are adopted to model ROIs in consecutive frames. Extensive experiments show that our method performs better than existing methods.