Research On Black Smoke Car Detection Technology Based On Video Image

Black smoke car intelligent monitoring system is very important for the current environmental protection。At present,the commonly used method is manual testing,which takes much time and effort.The intelligent monitoring system can maintain a long working time and can ensure the accuracy of test results,so it has a wide range of application prospects.Based on the monitoring video,this paper realizes the functions of car detection and tracking,vehicle tail feature extraction,training and testing a classifier of black smoke car and so on,the paper consists of the following aspects:(1)Vehicle Detection and Tracking Technology.According to the situation of black smoke with different shapes and sizes,ViBe algorithm is first studied.For the existence of ghost region in ViBe algorithm and its insensitivity to slight changes,an improved method is proposed,which is improved respectively for background initialization and sample set update.Through the improvement,the suspected black smoke in the video can be better detected,which is conducive to the further determination of the black car.Combined with the location and the size of the center of the target,the vehicles are tracked by using the information between adjacent frames,so as to determine the uniqueness of the car.The method can reduce the missed inspection and repeated detection.(2)Vehicle Tail Feature Extraction.For the detected vehicle,the region of interest is selected first,then the features such as color,wavelet and texture are extracted from the region.Due to the fact that the areas of interest generally contain a significant amount of black smoke,the smoky and non-smoky vehicles can be effectively distinguished by analyzing the features.(3)Classifier Training.Classifier training is carried out by using the extracted car tail features.Aiming at the characteristics that the correlation between different features is not large and the orders of magnitude are different,a BP neural network with multi-characteristic cascade is proposed.Separate the training of multiple features,and then output the results into a new vector training network,the final recognition results is obtained from the second level network.(4)Software Design and Implementation.Through the research of the above key algorithms,this paper designs a smoke car intelligent detection system,which realizes the vehicle detection and tracking based on video or surveillance camera,feature extraction,smoke car detection and other functions.Finally,the system’s functionality and performance were verified by using different videos.