Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....image. For the range data, a gray scale rendering of the CAD model is embedded within a pseudo colored rendering of the range image. This initial hypothesis has been automatically generated by two upstream algorithms described elsewhere. The first algorithm detects vehicles based on their color [13], and the second generates a vehicle and pose estimate based on the appearance of the occluding contour in a range image [11] We will use this example to illustrate how our iterative predict and match system is able to refine the coregistration estimate while detecting and accounting for the ....

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.

....and sensor features are presented and demonstrated on real data. This more geometrically precise 3D framework for ATR promises more accurate match measures, and consequently, more robust target identification. Since the associated computational demands are considerable, less expensive detection [ BDHR94 ] and hypothesis generation [ Bev92b ] algorithms are being used to provide focus of attention. Thus, it is assumed that the coregistration algorithms are being asked to rank and resolve quite specific hypotheses generated by up stream processing. An example of such a hypothesis: there is an M113 ....

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.

....information. The ATR algorithm being developed will use a three stage strategy [BHP94] First, a detection process suggests regions of interest within the image worth further consideration as possible targets. An innovation at this stage is the use of color as an additional detection queue [BDHR94] . The second stage extends LADAR probing techniques [BJLP92] to generate target type and target pose hypotheses. Finally, given object type and pose hypotheses, an error reduction approach will generate a best fit match between the sensor and model features [aJRB94] Using both multi sensor data ....

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages1619--1626,Los Altos, CA, November1994. ARPA, Morgan Kaufmann.

....coupling between feature prediction and matching: modifying the features expected to be visible as matching progresses. The algorithms presented here are being developed to perform final verification within a larger Automatic Target Recognition (ATR) system [BHP95] Thus, upstream color detection [BDHR94] and range boundary matching algorithms [Bev92] provide hypotheses indicating a specific target is at roughly the following position and orientation relative to the sensor platform. Consequently, the primary aim of the matching algorithm presented here is to reliably refine the pose estimate and ....

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.

....For asserting the attractiveness of buy writes [242] among many other data mining applications. 28 SREERAMA K. MURTHY ffl Image processing: For the interpretation of digital images in radiology [294] for recognizing 3 D objects [39] for high level vision [187] and outdoor image segmentation [40]. ffl Language processing: For medical text classification [212] for acquiring a statistical parser from a set of parsed sentences [229] ffl Law: For discovering knowledge in international conflict and conflict management databases, for the possible avoidance and termination of crises and wars ....

Shashi D. Buluswer and Bruce A. Draper. Non-parametric classification of pixels under varying illumination. SPIE: The Int. Society for Optical Eng., 2353:529--536, November 1994.

....green camouflage against green grass and brush. Color detection is relatively mature and has been integrated and demonstrated running on the Unmanned Ground Vehicle. The color detection effort is led by the University of Massachusetts. The general approach to target detection as laid out in [BDHR94] is based upon more general work on the use of learned multivariate decision trees in computer vision [DBU94] Hypothesis Generation Given regions of interest generated by the color detection process, or any other detection algorithm, this second stage hypothesizes what type or types of vehicles ....

....is summarized in the following sections. 4 Progress on Key Components 4. 1 Color Detection The essential elements of this work along with results on data collected by ourselves and Martin Marietta at Fort Carson [BPY94a] were reported in the previous Image Understanding Workshop Proceedings [BDHR94] Since this initial description of the color detection work, the following has been accomplished: 1. An improved way of coalescing individual pixel detections into ROIs has been implemented. 2. The algorithm has been successfully integrated with the other Reconnaissance, Surveillance and Target ....

[Article contains additional citation context not shown here]

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Non-parametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.

....color space. Another way of describing MDTs is that they fit a piecewise planar function to a decision surface in a 3 D feature space. It should be noted that other non parametric classifiers could also be used for this task, including back propagation neural networks. However, as discussed in [ 15 ] the decision surfaces for the apparent color of physical objects in an outdoor scene are well described as piecewise planar functions in 3 D, and MDTs are therefore appropriate. Neural networks search for decision functions in higher dimensionality function spaces, and therefore require more ....

....system for ATR has been formally evaluated on the Fort Carson data set, both by the authors and independently by Ted Yachik of Gilfillan Associates Inc. LGA) Over 100 color images of military targets taken on 35mm film and then digitized onto Kodak CD were used in this evaluation. In [ 15 ] the authors evaluated the system at both a pixel and regionof interest level. At the pixel level, it correctly identified target pixels 53.4 of the time and background pixels (which are much more common) 97.5 of the time, albeit with a high deviation from image to image. The SD was 10.4 ....

[Article contains additional citation context not shown here]

Shashi Buluswar, Bruce A. Draper, Allen Hanson, and Edward Riseman. Nonparametric Classification of Pixels Under Varying Outdoor Illumination. In Proceedings: Image Understanding Workshop, pages 1619--1626, Los Altos, CA, November 1994. ARPA, Morgan Kaufmann.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC