A.R. Hanson, E.M. Riseman, and C.A. Weems. Progress in computer vision at the university of massachusetts. In IUW, pages 39--47. Morgan Haufmann, April 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....He used Gabor wavelets as image features. Allen, Boult, Kender, and Nayar [1994] described work on learning object descriptions in eigenspaces. The objects are represented on a 20 dimensional manifold and recognition is done by matching new instances of object descriptions to these manifolds. Hanson, Riseman and Weiss [1994] described the Schema learning system. Their system consists of learning programs and image understanding routines. Fischler and Bolles [1994] described work on applying learning from experience to natural object recognition. Shafer, Kanade and Ikeuchi [1994] considered the problem of learning ....

Hanson, A.R., Riseman, E.M, and Weiss, R., "Progress in computer vision at the University of Massachusetts," Proceedings of the 1994 Image Understanding Workshop, pp. 43--51, 1994.

....were significant and calibration and a priori knowledge of the ground plane may not be useful in improving qualitative obstacle detection in this case. Fig. 7 is a sample of a sequence stereo picture of a real road scene with obstacles, taken by the stereo cameras onboard the Mobile Perception Lab[10]. The height of cameras, H, is 7.8 ft. Again, here we only show the left image for each stereo frame. EGP was run on this sequence. The first three image pairs Fig. 7(a) b) c) are used to give an initial estimate of the state vector which includes information about the ground plane. It is ....

A.R. Hanson, E.M. Riseman, and C.A. Weems. Progress in computer vision at the university of massachusetts. In IUW, pages 39--47. Morgan Haufmann, April 1993.

....case, though, the experimental result suggests that the available ground plane information and or camera calibration information might not be accurate enough. Fig. 9 is a sample of a stereo sequence of a real road scene with obstacles, taken by the stereo cameras onboard the Mobile Perception Lab[11]. The height of the cameras, H, is 7.8 ft. Again, only the left image is shown for each stereo frame. The two stereo cameras are mounted with parallel axes and a tilt angle of 11:8 ffi and a baseline of 2.55 ft. With this tilt angle, the look ahead distance range is from 32.2 ft. to 78 ft. EGP ....

A.R. Hanson, E.M. Riseman, and C.A. Weems. "Progress in computer vision at the university of massachusetts". In Proc. IUW, pages 39--47. Morgan Haufmann, April 1993.

....deviations of the points from a plane. All the three algorithms have been tested on real image data[9] Here we only report part of the experimental results using the EGP algorithm. Fig.1 is the left image of a sample of a sequence taken by the stereo cameras onboard the UMass Mobile Perception Lab[5]. The height of the cameras, h c , is 7.8 ft. The vehicle moved about 6 ft. between frames and the fifth frame was taken about 50 ft. away from the obstacles. The 6 points labeled in the image are the obstacle points. The heights of ground points 1 to 5 are 2.35 ft. and point 6 is 1.50 ft. For ....

A.R. Hanson, E.M. Riseman, and C.A. Weems. Progress in computer vision at the university of massachusetts. In IUW, pages 39--47. Morgan Haufmann, April 1993.

....3D pose computation step. The emphasis over the past year has been on improving both the reliability and efficiency of the search processes. The 3D pose refinement technique developed earlier works with either points or lines (Kumar 1989; Kumar and Hanson 1990a; Kumar and Hanson 1990b; Kumar and Hanson 1989a) and has been extended to be robust in the presence of outliers. The robustness is achieved at a computational cost, since the median of the error function is minimized by combinatorial methods over the subset space of all matched image and model lines. However, the method is capable of handling ....

.... sequences; the results are summarized in (Sawhney and Hanson 1990; Sawhney and Oliensis 1990) In addition, the results have been compared with two other motion algorithms: Adiv (Adiv 1985) Horn s relative orientation algorithm (Horn 1988) as well as Kumar s pose refinement algorithm (Kumar and Hanson 1989a; Kumar and Hanson 1989b) The results are preliminary and represent a continuing effort in understanding robust 3D reconstruction from monocular motion. More accurate results applied to a more varied data set awaits precise calibration of our cameras. 3. Interpretation of Static Scenes 3.1 ....

[Article contains additional citation context not shown here]

Hanson, A. and E. Riseman. (1989). "Progress in Computer Vision at the University of Massachusetts," Proc. of DARPA Image Understanding Workshop, Palo Alto, CA, pp. 49-55.

....of a site model of an area from multiple images (site images) taken over different days, from a variety of positions, and under variable weather and lighting conditions. Uses of the site model include 3D site visualization and familiarization, mission planning and assessment, and change detection [3, 9]. Consequently, an important component of the site model is an accurate geometric representation of significant objects in the site, such as buildings, as polygonal models with associated surface texture maps derived from the site images. Collins, et al. 5, 4] describe recent progress in RADIUS ....

E. Riseman, A. Hanson, R. Collins, B. Draper and R. Weiss, "Progress in Computer Vision at the University of Massachusetts," Proc. Arpa Image Understanding Workshop, Monterey, CA, 1994, pp. 43-51.

....ground height. All the three algorithms have been tested on real image data[10] Due to the limitation of space, we here only report part of one experimental results with EGP algorithms. Fig.1 is the left image of a sample of a sequence taken by the stereo cameras onboard our Mobile Perception Lab[6]. The height of the cameras, h c , is 7.8 ft. The 6 points labeled in the image are the obstacle points. The ground truth heights for points 1 to 5 are 2.35 ft. and for point 6 is 1.50 ft. Table 1 shows the estimated heights based on EGP algorithm for the 6 obstacle points. When more frames are ....

A.R. Hanson, E.M. Riseman, and C.A. Weems. Progress in computer vision at the university of massachusetts. In IUW, pages 39--47. Morgan Haufmann, April 1993.

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