Content-based Information Retrieval For Traffic Monitoring Videos

Nowadays, the application of video surveillance is increasing rapidly. It is ableto real-time monitor and record the situation as evidence of a certain area through theobservation and video capture froma certain camera. One of the most importantapplication areas of video surveillance is the traffic surveillance video informationretrieval, which has significant importance for traffic surveillance and control, is animportant part of the Intelligent Transportation Systems.But with the increase of surveillance cameras as well as the captured information,manually retrieval or identification will face enormous workload. Especially invehicle management and road monitoring system, where real-time video retrieval andidentification are badly needed to make the transportation planning andprocessing.How to make your computer automatically execute vehicle retrieval is animportant issue in the area of video surveillance.In this paper, aiming at the characteristics of the traffic monitoring video,researches are done on moving target extraction and image feature extractionmethod.First a rapid extraction of moving vehicles in the traffic monitoring video isproposed.And then feature extraction and similarity calculation is done to the vehicleto finally achieve the purpose of moving vehicles retrieve the video. Maincontributions of this paper are as follows:(1) A rapid extraction of moving vehicles in the traffic monitoring video isproposed and implemented. This algorithm uses hypothesis testing to higher orderstatistics of frame differences to achieve the rough separation of moving vehicles andbackground. Then obtain the length of the vehicle and extract the vertical coordinatesof the initial point of moving vehicle by setting a static scanning window with astationary location, combining with the velocity of the vehicle and the moving pixeldistribution probability in the window. And obtain the width of the vehicle and extractthe horizontal coordinates of the initial point of moving vehicle by setting a dynamicscanning window with an alterable location, combining with the distribution probability of moving pixels in the window. Finally this algorithm achieved the quickextraction of vehicles with the window obtained by length and width, combining withthe coordinates of the initial point of moving vehicle.(2) A retrieval algorithm combining the geometric feature and topological featureof the related vehicle image is proposed. This method first divide the target vehicleimage into five parts, then calculate theratio of geometric feature values between thefive parts to determine the geometric feature of the vehicle. Second it uses the dividedvehicle image to obtain the topological relations, and by calculating the ratio betweenthe length of the common boundary between the areas and length of total boundaries,to determine the vehicle topological features.Get the general feature of the vehicleimage bycombining the two features to. Finally, to get the similarity between thevehicle images, which are used for vehicle retrieval, by calculating the similaritybetween the areas.