Sub-Task 1-3. Image Processing and Understanding for Vision Applications

---

Progress  |  References  |  For more information

Background

A number of image processing techniques are key elements of applied vision research. In particular, image registration is a fundamental image processing technique, which has numerous applications to visual research, in addition the more traditional application domains such as remote sensing. Other image processing techniques, such as image compression, are also relevant to the handling of large numbers of images for visual research. Image Understanding refers to automated extraction of information from images. Our primary application to date has been automatic computation of gaze direction from images of the eye, in support of eye-movement research, and there are many applications for more general recognition of human gestures. For applied work in flight simulators, we have proposed registering images from a head-mounted scene camera to a stored model of the environment, in order to track the movement of the subject's head, while providing the experimenter with a pilot's eye view of the cockpit scene.

Objectives

This sub-task has four interrelated objectives:

1. Developments of algorithms, which are optimized for targeted applications in applied vision research.
2. Improvement of existing registration algorithms to handle registration of images of curved surfaces.
3. Application of registration techniques to recover head position from images from a head-mounted scene camera.
4. Optimization of image compression parameters for various machine vision tasks. Improvement of image processing methods for gaze tracking using images of the pupil.

Approach

Algorithms are developed and tested using empirical video data obtained from applied studies. Validation of the algorithms is done using simulated input sequences for which "ground truth" is known. Synthetic imagery is often based on real imagery in order to retain as many properties of the real images as possible.

Level 3 Milestones

FY98 Development of robust pupil tracking algorithm for head-mounted eye camera.
FY99 Development of head tracking algorithm for scene camera images.
FY00 Validation of distortion-fitting method for processing of SLO images.
FY01 Optimal compression parameters for machine vision tasks.
FY02 Refined temporal error diffusion algorithm.
FY03 Real-time implementations.

Point of Contact

Jeffrey B. Mulligan

(650) 604-3745
(650) 604-3323 (FAX)

jbm@vision.arc.nasa.gov