Research Of Automatic Detection Technology Of Abandoned Objects On Oil Field Roads

With the fast development of oil industry all over the world in recent years, the number of traffic vehicles on the special transport routes of the oil companies is increasing. And the problem of traffic safety is becoming more obvious. In such case, a system which can detect and recognize abandoned objects precisely is vital for traffic safety and enterprise production.There are several problems of the traditional method of abandoned objects detection. First, because most of the sections mainly rely on artificial detection, the negligence of employees will easily cause detection accidents. Second, the slow response speed of artificial detection can’t deal with the problem timely, so the accident frequency is much higher. Third, the workload is becoming heavier because the number of special roads paved by the oil companies is increasing. For these problems such as low detection accuracy; slow response speed and large workload, more reasonable methods and more advanced technology are needed to complete the job of abandoned object detection on oilfield roads.This thesis deeply researches the feature detection and recognition of abandoned objects according to the actual situation of Daqing oil field roads. The main contributions of this thesis are as follows.First, a road background model is set up which has good anti-interference performance and can extract the foreground image away accurately. The thesis uses a combination of Gaussian Mixture Model and Background Subtraction algorithm which based on the specificity of oil field roads to build reliable background model.Second, the thesis processes the abandoned objects, including shadow elimination, features extraction and matching. It uses SURF operators to detect and describe the features of abandoned objects while building feature dictionary. Then it uses the algorithm of FANN which has already improved to generate feature vector by matching the feature points automatically. The FANN algorithm is also used to complete the code book building of BOW model and the training of the samples at the same time. All these steps are based on the principle of BOW model.Third, due to the problems of KNN classifier such as slow detective speed and low recognition accuracy, the thesis uses ISOMAP algorithm to reduce the dimension of sample space in which way it can improve the computing speed of the classifier further. It also uses the text weighted algorithm of TF-IDF to improve recognition accuracy of the classifier.Fourth, we develop the system of automatic detection of abandoned objects on oil field roads according to the algorithm proposed by this thesis. The actual application of the system shows that the detection and recognition method of abandoned objects can better adapt to the situation of oil field roads. The system is not only easy to operate, but also performs faster. Besides, it has higher recognition efficiency. The actual application effect is more ideal.