Christopher K. Eveland

Background modeling for segmentation of video-rate stereo sequences

Stereo sequences promise to be a powerful method for segmenting images for applications such as tracking human figures. We present a method of statistical background modeling for stereo sequences that improves the reliability and sensitivity of segmentation in the presence of object clutter. The dynamic version of the method, called gated background adaptation, can reliably learn background statistics in the presence of corrupting foreground motion. The method has been used with a simple head discriminator to detect and track people using a stereo head mounted on a pan/tilt platform. It runs at video rates using standard PC hardware.

Illumination invariant face recognition using thermal infrared imagery

A key problem for face recognition has been accurate identification under variable illumination conditions. Conventional video cameras sense reflected light so that image grayvalues are a product of both intrinsic skin reflectivity and external incident illumination, thus obfuscating the intrinsic reflectivity of skin. Thermal emission from skin, on the other hand, is an intrinsic measurement that can be isolated from external illumination. We examine the invariance of Long-Wave InfraRed (LWIR) imagery with respect to different illumination conditions from the viewpoint of performance comparisons of two well-known face recognition algorithms applied to LWIR and visible imagery. We develop rigourous data collection protocols that formalize face recognition analysis for computer vision in the thermal IR.

Tracking human faces in infrared video

Abstract Detection and tracking of face regions in image sequences has applications to important problems such as face recognition, human–computer interaction, and video surveillance. Visible sensors have inherent limitations in solving this task, such as the need for sufficient and specific lighting conditions, as well as sensitivity to variations in skin color. Thermal infrared (IR) imaging sensors image emitted light, not reflected light, and therefore do not have these limitations, providing a 24-h, 365-day capability while also being more robust to variations in the appearance of individuals. In this paper, we present a system for tracking human heads that has three components. First, a method for modeling thermal emission from human skin that can be used for the purpose of segmenting and detecting faces and other exposed skin regions in IR imagery is presented. Second, the segmentation model is applied to the CONDENSATION algorithm for tracking the head regions over time. This includes a new observation density that is motivated by the segmentation results. Finally, we examine how to use the tracking results to refine the segmentation estimate.

Quantitative measurement of illumination invariance for face recognition using thermal infrared imagery

A key issue for face recognition has been accurate identification under variable illumination conditions. Conventional video cameras sense reflected light so that image gray values are a product of both intrinsic skin reflectivity and external incident illumination, obfuscating intrinsic reflectivity of skin. It has been qualitatively observed that thermal imagery of human faces is invariant to changes in indoor and outdoor illumination, although there never has been any rigorous quantitative analysis to confirm this assertion published in the open literature. Given the significant potential improvement to the performance of face recognition algorithms using thermal IR imagery, it is important ot quantify observed illumination invariance and to establish a solid physical basis for this phenomenon. Image measurements are presented from two of the primarily used spectral regions for thermal IR; 3-5 micron MidWave IR and the 8-14 micron LWIR. All image measurements are made with respect to precise blackbody ground-truth. Radiometric calibration procedures for two different kinds of thermal IR sensors are presented and are emphasized as being an integral part to data collection protocols and face recognition algorithms.

Face Recognition in the Thermal Infrared

Recent research has demonstrated distinct advantages of using thermal infrared imaging for improving face recognition performance. While conventional video cameras sense reflected light, thermal infrared cameras primarily measure emitted radiation from objects such as faces. Visible and thermal infrared image data collections of frontal faces have been on-going at NIST for over two years, producing the most comprehensive face database known to involve thermal infrared imagery. Rigorous experimentation with this database has revealed consistently superior recognition performance of algorithms when applied to thermal infrared, particularly under variable illumination conditions. Physical phenomenology responsible for this observation is analyzed. An end-to-end face recognition system incorporating simultaneous coregistered thermal infrared and visible has been developed and tested indoors with good performance.

Using infrared sensor technology for face recognition and human identification

Recent research has demonstrated distinct advantages using thermal infrared imaging for improving face recognition performance. While conventional video cameras sense reflected light, thermal infrared cameras primarily measure emitted radiation from objects at just above room temperature (e.g., faces). Visible and thermal infrared image data collections of frontal views of faces have been on-going at NIST for over two years producing the most comprehensive database known to involve thermal infrared imagery of human faces. Rigorous experimentation with this database has revealed consistently superior recognition performance of algorithms when applied to thermal infrared particularly under variable illumination conditions. An end-to-end face recognition system incorporating simultaneous coregistered thermal infrared and visible has been developed and tested both indoors and outdoors with good performance.

Particle Filtering with Evidential Reasoning

Particle filtering has come into favor in the computer vision community with the CONDENSATION algorithm. Perhaps the main reason for this is that it relaxes many of the assumptions made with other tracking algorithms, such as the Kalman filter. It still places a strong requirement on the ability to model the observations and dynamics of the systems with conditional probabilities. In practice these may be hard to measure precisely, especially in situations where multiple sensors are used. Here, a particle filtering algorithm which uses evidential reasoning is presented, which relaxes the need to be able to precisely model observations, and also provides an explicit model of ignorance.

Versatile low-power multispectral video fusion hardware

Image fusion of complementary broadband spectral modalities has been extensively studied for providing performance enhancements to various military applications. With the growing availability of COTS and customized video cameras that image in VIS-NIR, SWIR, MWIR and LWIR, there is a corresponding increase in the practical exploitation of different combinations of fusion between any of these respective spectrums. Equinox Corporation has been developing a unique line of products around the concept of a single unified video image fusion device that can centrally interface with a variety of input cameras and output displays, together with a suite of algorithms that support image fusion across the diversity of possible combinations of these imaging modalities. These devices are small in size, lightweight and have power consumption in the vicinity of 1.5 Watts making them easy to integrate into portable systems.

Image fusion of shortwave infrared (SWIR) and visible for detection of mines, obstacles, and camouflage

Over the last decade there has been study of separating ground objects from background using multispectral imagery in the reflective spectrum from 400-2500nm. In this paper we explore using two broadband spectral modalities; visible and ShortWave InfraRed (SWIR), for detection of minelike objects, obstacles and camouflage. Whereas multispectral imagery is sensed over multiple narrowband wavelengths, sensing over two broadband spectrums has the advantage of increased signal rsulting from integrated energy over larger spectrums. Preliminary results presented here show that very basic image fusion processing applied to visible and SWIR imagery produces reasonable illumination invariant segmentation of objects against background. This suggests the use of a simplified compact camera architecture using visible and SWIR sensing focal plane arrays for performing detection of mines and other important objects of interest.