Object tracking from compressed video using Kalman Filter and a novel spatiotemporal motion-vector filter.

• Main Author: Ronaldo Carvalho Moura Júnior
• Publication Date: 2010
• Format: Master thesis
• Language: eng
• Source: Biblioteca Digital de Teses e Dissertações do ITA
• Download full: http://www.bd.bibl.ita.br/tde_busca/arquivo.php?codArquivo=1070
• topic: Processamento digital de sinais; Filtros de rastreamento; Sinais de vídeo; Compressão de vídeo; Estimação de sistemas; Compensação de movimento de imagens; Visão por computadores; Telecomunicações; Engenharia eletrônica

Video Object Tracking plays a crucial role on several Computer Vision applications, such as Video Surveillance, Intelligent Transportation System (ITS), Human Machine Interface (HMI), Video Indexing and Shopping Behavior Analysis. Nevertheless, the processing power demanded by object tracking techniques still consists in a bottleneck to their wider adoption. To reduce this computational power demand, some techniques that extract object motion information from compressed video domain, instead of the raw video, have been developed. This work addresses the problem of efficiently tracking objects from compressed video. The focus is on algorithms that track objects using motion estimation informationfrom MPEG-2 and MPEG-4 family of video compressors. Two complementary solutions are presented. At first, a novel Spatiotemporal Motion-Vector Consistency Filter is proposed and evaluated. The filter is applied on the initial stage of tracking algorithm and significantly reduces the noisy motion vectors which do not represent a real object movement. Then, a Kalman Filter is used to provide improved estimations of objects position and size. A novel model for Kalman Filter application on the context of motion-vector based object tracking is proposed and evaluated, with determination of measures and noise patterns. Qualitative and quantitative experiments, with standard metrics, are performed displaying that the proposed Spatiotemporal Filter outperforms the currently widely used Vector Median Filter. The results obtained with the Spatiotemporal Filter make it suitable as a first step of any system that aims to detect and track objects from compressed video using its motion vectors. Both filters are jointly used in a complete object tracker system denominated moveTRAKS - motion-vector based object Tracker with Kalman filter and Spatiotemporal filter. The moveTRAKS is also qualitative and quantitative tested, demonstrating its efficiency and limitations for compressed video object tracking.