A similarity analysis and clustering algorithm for video based on moving trajectory time series wavelet transform of moving object in Video

Time-series data mining is a very important element of data mining. As a typical time-series data, video data has been widely used for many applications such as film, medical, sports and traffic areas. In this paper we integrates several technologies into the video moving object time-series data mining. We propose a similarity analysis and clustering algorithm for videos based on the moving trajectory time series data wavelet transform of the moving object in the videos. Utilizing the image difference technique, our algorithm detects the moving objects from the scene surveillance video, calculates the centroid of the moving object, and uses the centroid series to character the moving trajectory of the object. Then we use wavelet analysis method to achieve dimensionality reduction and get the first k wavelet coefficient to substitute the original motion time-serial data. Based on the Euclidean distance, we utilize two judgement rules to determine the similarity of time-serial data and cluster them, use the rules to perform the similarity search and clustering of video. We apply the algorithm to athletic sports videos analysis.