Segmentation

This paper presents an overview of Intelligent Video work currently under development at the GE Global Research Center and other research institutes. The image formation process is discussed in terms of illumination, methods for automatic camera calibration and lessons learned from machine vision. A variety of approaches for person detection are presented. Crowd segmentation methods enabling the tracking of individuals through dense environments such as retail and mass transit sites are discussed. It is shown how signature generation based on gross appearance can be used to reacquire targets as they leave and enter disjoint fields of view. Camera calibration information is used to further constrain the detection of people and to synthesize a top-view, which fuses all camera views into a composite representation. It is shown how site-wide tracking can be performed in this unified framework. Human faces are an important feature as both a biometric identifier and as a method for determining the focus of attention via head pose estimation. It is shown how automatic pantiltzoom control; active shape/appearance models and super-resolution methods can be used to enhance the face capture and analysis problem. A discussion of additional features that can be used for inferring intent is given. These include body-part motion cues and physiological phenomena such as thermal images of the face.

This paper presents an overview of Intelligent Video work currently under development at the GE Global Research Center and other research institutes. The image formation process is discussed in terms of illumination, methods for automatic camera calibration and lessons learned from machine vision. A variety of approaches for person detection are presented. Crowd segmentation methods enabling the tracking of individuals through dense environments such as retail and mass transit sites are discussed. It is shown how signature generation based on gross appearance can be used to reacquire targets as they leave and enter disjoint fields of view. Camera calibration information is used to further constrain the detection of people and to synthesize a top-view, which fuses all camera views into a composite representation. It is shown how site-wide tracking can be performed in this unified framework. Human faces are an important feature as both a biometric identifier and as a method for determining the focus of attention via head pose estimation. It is shown how automatic pantilt-zoom control; active shape/appearance models and super-resolution methods can be used to enhance the face capture and analysis problem. A discussion of additional features that can be used for inferring intent is given. These include body-part motion cues and physiological phenomena such as thermal images of the face.

The goal of this work is to provide information for the automatic control of an active camera. One way to achieve this is to run a set of saliency detectors in the scene which is tuned to a specific class of objects. As opposed to this, we propose a formal definition for what we mean for salient motion by means of the concept of stationarity of stochastic processes. We propose a hierarchy of generative models for salient and nuisance motion, along with recursive learning procedures for on-line processing. By doing so, we turn the problem of detecting salient motion into the problem of model change detection, which we solve optimally online using the sequential generalized likelihood ratio test. Our system is designed for realtime applications, and is able to detect salient motion even in severely cluttered scenes.

We present a novel approach to improving moving object detection algorithms for aerial video by incorporating scene understanding. We argue that scene understanding is a valuable input feature for moving object segmentation, and show how it can be used to reduce the false positive rate of any moving object detection algorithm. The approach is especially effective on sequences with large scene depth and much parallax, as often occurs with low, fast-moving sensors. In one difficult sequence, our results show an 8-fold reduction of false positives on average, with essentially no impact on the true positive rate, even when scene understanding is only moderately accurate. We show how scene understanding can be used to increase the accuracy of frame-to-frame homography estimates, and also present a novel approach that uses motion information from the previous frame for outlier rejection at the current frame to further increase accuracy.