Data Mining Framework for Surveillance Video

This paper presents a framework for surveillance videos of stationery places. To start with, we implement an algorithm to group incoming video stream into meaningful pieces called segments. Further we extract a feature of segment (i.e. motion) which is used to characterize the segments. Motion of a segment is extracted using two dimensional matrix which is constructed using accumulated pixel differences among all frames in a segment. Video segments are then clustered using K-means algorithm. Then we find the abnormality in the segments of the video. 1. INTRODUCTION Video surveillance systems are now used at various places like banks, hotels, schools and so on. These systems are used for real-time monitoring or post checking. Current video surveillance systems have lower intelligence and it need people to monitor them. So itis required to have some intelligence within systems so that systems can take decision on their own without need of people to monitor them[5, 6]. Data mining is very active research topic nowadays and is the process of extracting previous knowledge and detecting interesting patterns from large set of data. There are various tools and product available in market for video surveillance. Among them traffic surveillance systems [3, 4] are getting more popular because of the use and decision making of the system and there is availability of cheap sensors and processors at reasonable cost. Most of the video surveillance systems provides motion detection and there is facility for recording video when motion is detected so as to reduce the storage and processing time and It can also record a video in different formats and allows multiple cameras to be operated and provides view of multiple cameras at a single place and produces reasonable good results. The speed and amount of motion in surveillance video can reflects the amount of moving targets and can also model the motion due to stationery surveillance camera and motion which comes from objects in surveillance video. The video is divided into frames and we group incoming video frames into meaningful pieces called segments. For each of the segment, feature (motion)is extracted. Segments are further