Target Detection And Subtraction In Video Survielliance System

With the development of computer, network, video transmission and artificial intelligence, surveillance video has been widely used in various fields, such as security, transportation and other situations. Abnormal events detecting is an important and challenging task in surveillance video applications. In surveillance video applications, temporarily static regions indicate move-then-stop objects, such as the abandoned/removed objects, parked vehicles. This paper presents an approach for the detection of move-then-stop abnormal objects.The proposed approach uses both pixel-level and region-level analysis together. In the pixel-level foreground extraction process. In the first part, Gaussian background model is introduced. According to the characteristics of target objects, we improved the background model to keep the target region stay in foreground segmentation results. The dual foreground concept based designed procedures could effectively detect and extract the residence pixels covered by target objects. In the second part, the original imaged is divided into superpixels first. With the pixel-level results and the superpixel segmentation results, a classifier is build to separate superpixels belonging to target from non-target ones. Since the intra-frame spatial correlation is used, the output result could be more accurate.The experiments are carried out on several standard surveillance video data sets. In this work, the experimental results are provided in3ways, the result images, the recall and accuracy of object extraction and the detection rate of abnormal events. Experimental results show that the proposed method is available in multi situations, and it could effectively improve the accuracy of object extraction, with higher detection rate and relatively low false alarm rate. From the result images it can be seen that the output result could fit the contour of target objects well.