Motion detection can play an important role in many vision tasks. Yet image motion can arise from "uninteresting" events as well as interesting ones. In this paper, salient motion is defined as motion that is likely to result from a typical surveillance target (e.g., a person or vehicle traveling with a sense of direction through a scene) as opposed to other distracting motions (e.g., the scintillation of specularities on water, the oscillation of vegetation in the wind). We propose an algorithm for detecting this salient motion that is based on intermediate-stage vision integration of optical flow. Empirical results are presented that illustrate the applicability of the proposed methods to real-world video. Unlike many motion detection schemes, no knowledge about expected object size or shape is necessary for rejecting the distracting motion.

Two important problems in camera control are how to keep a moving camera fixated on a target point, and how to precisely aim a camera, whose approximate pose is known, towards a given 3D position. This paper describes how electronic image alignment techniques can be used to solve these problems, as well as provide other benefits such as stabilized video. Hence, stabilized, fixated imagery is obtained despite large latencies in the control loop, even for simple control strategies. These techniques have been tested using an airborne camera and real-time affine image alignment.

We use Sarnoff's VFE-200 next-generation video processor to demonstrate real-time algorithms for camera fixation, mosaic construction, and moving object detection and tracking from a moving camera. The algorithms are all derived from our ability to perform real-time image alignment. These capabilities are currently being used in several unmanned aerial and ground surveillance programs.

Vision systems that must operate autonomously over varying environmental conditions often must use different parameter values or algorithms depending on these conditions. A key problem is how to automatically assess the incoming imagery to determine these appropriate parameters and algorithms. This paper presents methods for such assessment. Specifically, it presents measures for determining whether the scene is well-lit (i.e. whether objects entire extent is visible, versus just their lights), whether the scene has sufficient contrast, and whether objects are casting shadows. The methods are applied in the domain of traffic monitoring, are based on empirical data, and have been tested on videotape segments from over 25 different scenes.

A key to developing computationally efficient stereo vision is the incorporation of intelligent control. Stereo is most effective when it is able to "focus" its analysis on regions and details of a scene that are important to the task at hand, while avoiding less important regions and unnecessary detail. The paper describes two methods for electronically "focusing" stereo measurement through simple image pre-processing. The first allows measurement sensitivity to be adjusted. The second allows the shape of the 3D region in which measurements are gathered to be matched to the shape of surfaces in the scene.>

Searching for an object in a cluttered area is often complicated by the fact that the clutter hides portions of the area from view. Multiple viewpoints are often required. Previous viewpoint selection algorithms rely on models that represent the portions already viewed. Unfortunately, these models are often difficult to construct when the sensor position is imprecisely known and when the obtainable depth data is sparse. This paper questions whether scene models are really necessary for selecting viewpoints for search tasks. It investigates two alternative viewpoint selection methods that do not rely on such models. One constructs no representation and simply uses a fitted set of viewpoints, while the other selects viewpoints based on detected occluding edges. On average, both model-free methods compare very favorably to model-based methods, although additional work is necessary to ensure that they abort in a timely manner when the desired object cannot be found.>