Refine
Has Fulltext
- no (14)
Year of publication
Document Type
- Conference Proceeding (13)
- Article (1)
Language
- English (14)
Institute
In the past decades most object recognition systems were based on passive approaches. But in the last few years a lot of research was done in the field of active object recognition. In this context there are several unique problems to be solved. One of them is how to fuse a series of images that might differ in their viewpoints. In this paper we present a well-founded approach for the fusion of multiple views based on a recursive density propagation method. It uses particle filters for solving the fusion in a continuous pose space. Furthermore we will show by means of a statistical object recognition system how to integrate such systems into our fusion approach. The experimental result will show, how the fusion can improve classification rates substantial, es-pecially for difficult conditions like heterogeneous background within real world environments.
Improving Statistical Object Recognition Approaches by a Parameterization of Normal Distributions
(2004)
As statistical approaches play an important role in object recognition, we present a novel approach which is based on object mod- els consisting of normal distributions for each training image. We show how to parameterize the mean vector and covariance matrix independently from the interpolation technique and formulate the classification and localization as a continuous optimization problem. This enables the computation of object poses which have never been seen during training. For interpolation, we present four different techniques which are compared in an experiment with real images. The results show the benefits of our method both in classification rate and pose estimation accuracy.