Covariance Tracking
The goal of this project is to develop robust, fast, and highly discriminative object representations and to integrate these representations within a state-of-science tracker. We also aim to develop more accurate object detectors by using this descriptor. Object tracking is one the most important and challenging tasks in computer vision. It has commercially valuable applications ranging from video surveillance to traffic management to to video summarization and compression. Covariance tracking is robust against severe illumination changes, noise, and erratic motion. It has been presented in Japan in January 2006 press release and it received significant media attention. Several companies already expressed their interest on this technology and investigating the possibility of joint developments.
Background & Objective: Our objective is to improve the robustness and adaptability of surveillance tracking systems. We previously developed several tracking methods that have been already integrated into key products. Covariance based representation will increase the accuracy of these methods for uncontrolled lighting conditions and setups. Tracking has a wide spectrum of applications. For instance, in video surveillance, tracking assists understanding the movement patterns of people to uncover suspicious events. It is a key technology in traffic management to estimate flux and congestion statistics. Advanced vehicle control systems depend on the tracking information to keep the vehicle in lane and prevent from collisions. In robotics, tracking bridges the gap between the raw visual information and environmental awareness. In video summarization, it is applied to generate object-based representations and automatic content annotations. Tracking is also a fundamental technology to extract regions of interest and video object layers as defined in JPEG-2000 and MPEG-4 standards.
Technical Discussion: Covariance matrix representation embodies both spatial and statistical properties of objects, and provides an elegant solution to fusion of multiple features. Covariance is an essential measure of how much the deviation of two or more variables or processes match. In tracking, these variables correspond to point features such as coordinate, color, gradient, orientation, and filter responses. This representation has much lower dimensionality than histograms. It is robust against noise and lighting changes. To track objects using covariance descriptor, an eigenvector based distance metric is adapted to compare the matrices of object and candidate regions. Covariance tracker does not make any assumption on the motion. This means that it can keep track of objects even if their motion is erratic and fast. It can compare any regions without being restricted to a constant window size. In spite of these advantages, the computation of the covariance matrix distance for all candidate regions is slow and requires exponential time. An integral image based algorithm that requires constant time is proposed to improve the speed.
Publications:
| Technical Reports: | |
| Covariance Tracking using Model Update Based on Lie Algebra | |
Technology Area: Computer Vision
Modification Date: September 12, 2007