| Black smoke exhaust not only affects air quality,but also endangers human health.It is of great significance to control smoky vehicles with high emission and high pollution on the road.Compared with the traditional manual screening and sensor monitoring methods,the intelligent recognition method of smoky vehicles based on computer vision can effectively reduce the consumption of human and material resources,maintain the objectivity and impartiality of the recognition process,and has broad application prospects.This thesis focus on the location algorithm of smoke exhaust area at the tail of intelligent smoky vehicles and the black smoke multi-feature fusion algorithm.The specific work is as follows.In order to reduce the influence of interference information such as road surface and vehicles body on feature extraction,the three frame difference method and progressive probabilistic hough transform method are combined to locate and improve the smoke exhaust area of interest at the rear of the vehicles,and add constraints,screen according to the inclination angle and position information of the detected edge contour line,retain the edge contour line at the bottom of the rear of the vehicles,and then determine the smoke exhaust area.Through the verification of the positioning effect,90% of the vehicles smoke exhaust area can be located effectively.Black smoke has the characteristics that its color is close to that of the road surface and its shape is not fixed.It is sensitive to the changes of environmental factors such as road surface and light.In order to extract effective black smoke features,a feature extraction algorithm based on feature fusion is proposed.According to the scale space theory,based on the Local Binary Pattern(LBP)features,Gaussian smoothing and down sampling are combined to fuse the local texture structure information of different scales of the smoke image,so as to optimize the texture expression ability in the spatial domain.The multi-scale and multi-directional transform domain features of black smoke based on Gabor are extracted,and the black smoke features are compressed and encoded by meshing and maximum suppression methods to filter out irrelevant interference information.Two heterogeneous features are fused to further expand the texture features in local spatial domain.The experimental results show that the algorithm can achieve the detection rate of 88.93% and the false positive rate of 5.97%,and the effect of black smoke recognition is obvious. |
