Arbel, T., Whaite, P. & Ferrie, F. P. (1994b), Recognizing volumetric objects in the presence of uncertainty, Technical Report TR-CIM-94-03, Center for Intelligent Machines, McGill University, Montr'eal, Qu'ebec, Canada. Available via ftp at ftp.cim.mcgill.ca in /pub/3d/papers/tr-cim-94-04.ps.gz.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Carlo simulation) to approximate the marginal distributions. Madsen [56] in studying the effect of viewpoint variation on pose estimation, considers the empirical and predictive variances of the pose parameters, but again fails to address their effects on point estimation. Arbel and Ferrie [1] use statistical estimation model densities for volumetric models, in a recognition by part approach. Also authors from the neural network community have been active in this field. Chandrasekaran et al. 17] make use of a Kohonen self organizing map to formulate class label prediction ....

T. Arbel. Recognizing volumetric objects in the presence of uncertainty. Master's thesis, McGill University, Montr'eal, Qu'ebec, Canada, April 1995.

No context found.

Arbel, T., Whaite, P. & Ferrie, F. P. (1994b), Recognizing volumetric objects in the presence of uncertainty, Technical Report TR-CIM-94-03, Center for Intelligent Machines, McGill University, Montr'eal, Qu'ebec, Canada. Available via ftp at ftp.cim.mcgill.ca in /pub/3d/papers/tr-cim-94-04.ps.gz.

No context found.

Arbel, T., Whaite, P. & Ferrie, F. P. (1994a), Recognizing volumetric objects in the presence of uncertainty, in `Proceedings 12th International Conference on Pattern Recognition', IEEE Computer Society Press, Jerusalem, Israel, pp. 470--476.

....they can represent as well as their computational simplicity. We have developed a method of avoiding degeneracies in the case of the superellipsoid, which permits the use of this convenient parametric form without incurring undue computational overhead. A discussion of this method can be found in [1]. 1 The extension of our recognition strategy to multi part objects is currently being investigated in our laboratory. 3 PARAMETRIC SHAPE RECOGNITION USING A PROBABILISTIC INVERSE THEORY We begin in Section 2 with a brief overview of the inverse theory [13] and then formulate the problem of ....

....superquadrics pose a number of problems due to degeneracies in shape and orientation. In solving this problem, we have represented objects by multi modal distributions, where each mode contains information about a possible equivalent form. A lengthy discussion of these techniques can be found in [1]. 3. Experiments Six objects were chosen for the purposes of testing the recognition procedure. These objects included a small wooden sphere (rad = 20mm) a slightly larger wooden sphere (rad = 25mm) a wooden block, a wooden cylinder, a plastic lemon, and a rounded block made of a sponge like ....

T. Arbel. Recognizing volumetric objects in the presence of uncertainty. Master's thesis, McGill University, Montr'eal, Qu'ebec, Canada, April 1995.

....how uncertainty conditions prior expectations such that the shape of the resulting belief distribution can vary greatly, becoming very delta like as the interpretation tends towards certainty. In contrast, ambiguous or poor interpretations consistently tend towards very broad or flat distributions [4]. We exploit this characteristic to define the notion of an informative viewpoint, i.e. a view which gives rise to assertions that have a high probability according to their associated belief distribution. There are at least two applications for this result. First, in the case of an active ....

.... by Tarantola in [25] Earlier work has shown how this theory can be used to methodically synthesize belief distributions corresponding to each model hypothesis, H i , given the parameters corresponding to the unknown model, M, computed from the current measurement D j , i.e. P (H i jMD j ) [4, 3]. This procedure explicitly accounts for uncertainties arising from the estimation of the unknown model parameters, database model parameters, and prior expectations on the frequency of occurrence for each of the database entries. In this case, the solution reduces to the classical Bayesian ....

[Article contains additional citation context not shown here]

T. Arbel, P. Whaite, and F. P. Ferrie. Recognizing volumetric objects in the presence of uncertainty. In Proceedings 12th International Conference on Pattern Recognition, pages 470--476, Jerusalem, Israel, Oct 9-13 1994. IEEE Computer Society Press.

....in viewpoint conditions the respective belief distributions, as they are normalized with respect to a global frame of reference. As a result, relative values between the views are meaningless. The normalizing factor is some unknown function of viewpoint, and is difficult to obtain analytically. In [2], it is indicated that the reason for the difficulty lies in that this factor can be shown to be a function of R D dd, the volume of data space. The issue of how to define this space is a difficult one to address. In order to do so, a commitment to a permissible region of observed parameters ....

T. Arbel. Recognizing volumetric objects in the presence of uncertainty. Master's thesis, McGill University, Montr'eal, Qu'ebec, Canada, April 1995.

.... This very line of thought has already lead to investigations in the area of autonomous exploration for shape reconstruction [25, 26] Although most of what follows does not strictly depend on a particular experimental set up, the measurement system we used is the autonomous explorer described in [1, 26]. We argue that learning 2 the recognition task directly from the shape recovery data can provide relevant insight to the above issues. As in recent work on the same subject [1] the recovery system itself is regarded as an erroneous measurement device, and the stress herein is on how to ....

....on a particular experimental set up, the measurement system we used is the autonomous explorer described in [1, 26] We argue that learning 2 the recognition task directly from the shape recovery data can provide relevant insight to the above issues. As in recent work on the same subject [1], the recovery system itself is regarded as an erroneous measurement device, and the stress herein is on how to properly use the estimated uncertainty in the recovered parametric shapes for the recognition purposes. However, in the former approach the parameters to object match is learned by ....

[Article contains additional citation context not shown here]

T. Arbel, P. Whaite, and F. P. Ferrie. Recognizing volumetric objects in the presence of uncertainty. In Procedings 12th International Conference on Pattern Recognition, Jerusalem, Oct. 1994. IEEE Computer Society Press.

....the smaller sphere and the lemon was then tested. The result is the belief distribution found in Figure 2c. One can see that the system has a significantly higher degree of confidence in the hypothesis that the measured model was a large sphere. 3. Determining Which Viewpoints are Informative In [3], it has been shown that recognition based on complete information produced perfect results in all cases. Since complete information is not always available, and potentially expensive to acquire, recognition schemes based on single viewpoints are required. We will show that recognition based on ....

....5.1. Characterizing Informative Viewpoints by External Threshold. It has already been shown that the system can successfully recognize an instance of any object in the database with perfect results, provided that all its surfaces are accessible, independently of viewpoint and sampling order [3]. It has also been shown that recognition from single views retains most of the selectivity of the previous case, displaying high discriminatory powers, with the occasional false positive cases. The hypothesis has been that, if the majority of the incorrect cases occur with low beliefs, by raising ....

T. Arbel, P. Whaite, and F. P. Ferrie. Recognizing volumetric objects in the presence of uncertainty. In Proceedings 12th International Conference on Pattern Recognition, pages 470-- 476, Jerusalem, Israel, Oct 9-13 1994. IEEE Computer Society Press.

....serves to condition prior expectations such that the shape of the resulting belief distribution can vary greatly, becoming very delta like as the interpretation tends towards certainty. In contrast, ambiguous or poor interpretations consistently tend towards very broad or flat distributions [2]. We exploit this characteristic to define the notion of an informative viewpoint, i.e. a view which gives rise to assertions that have a high probability according to their associated belief distribution. There are at least two applications for this result. First, in the case of an active ....

.... by Tarantola in [24] Earlier work has shown how this theory can be used to methodically synthesize belief distributions corresponding to each model hypothesis, H i , given the parameters corresponding to the unknown model, M, computed from the current measurement D j , i.e. P (H i jMD j ) [2]. This procedure explicitly accounts for uncertainties arising from the estimation of the unknown model parameters, database model parameters, and prior expectations on the frequency of occurrence for each of the database entries. In this case the solution reduces to the classical Bayesian ....

T. Arbel, P. Whaite, and F. P. Ferrie. Recognizing volumetric objects in the presence of uncertainty. Technical Report TR-CIM-94-03, Center for Intelligent Machines, McGill References 25 University, Montr'eal, Qu'ebec, Canada, Mar. 1994. Available via ftp at ftp.cim.mcgill.ca in /pub/3d/papers/tr-cim-94-04.ps.gz.

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