Global visual localization of mobile robots using kernel principal component analysis

Abstract

The aim of this article is to present the potential of Kernel Principal Component Analysis (Kernel PCA) in the field of vision based robot localization. Using Kernel PCA we can extract features from the visual scene of a mobile robot The analysis Is applied only to loel fealum so as to guarantee better computational performance as well as translation invariance. Compared with the classical Prinripal Component Analysis (PCA), Kernel PCA results show superiority in localization and robustness in presence of noisy scenes. The key success of the kernel PCA is the use of fractional power polynomial kernels. I. INTRODUCTlUNThe problem of robot localization can be classified as either global or local localization [IO]. In global localization, the robot tries to discover its position without previous knowledge about its location. In local localization, the robot must update its position using its current data from its sensors as well as the previous information that it has already accumulated. The lack of any historical information about its surroundings makes the global localization more challenging [7]. The idea of feature based robot localization involves representing the robot environment as a topological map by means of a large set of features. The main properties of these features are: (1) They should be a compressed form of the original Scenes so as to speed up the computation of the comparisons, still, they should maintain distinguishing representations of the scenes. (2) They should exhibit invariance against different transformations on the scenes such as translation and scale. (3) They should also exhibit robusmess against noise or illumination changes, which the robot encounters during its navigation. PCA has been applied in the field of robot localization. In [SI active vision is combined with robot localization using PCA. In [I] and [I31 the study of the problem of batch learning and the use of incremental PCA is presented.Their idea is lo deal with on-line leaning of the robot landmarks without recomputing the PCA for the whole samples each time. The work done in [41 presents the effect of illumination on PCA. They present illumination invariant feaNres by filtering the eigenimages rather than filtering the original samples. In [12] a comparison among different vision-based mbot localization approaches is made. Their results show that PCA is more robust and accurate than other methods such as edge density based, hut also show that PCA requires more computation power. Robot localization using PCA can be classified as either local and global based on the feature extraction applied. In the global based approach [8], the whole image is considered as a sample and applied to the PCA as a vector. An example of global features is the work done in [31, where PCA is globally applied to panoramic images, hey introduce robust PCA using an expectation maximization approach where outliers can be resolved. On the other band, in the local based approach, a set of landmarks (small patches) are first selected from the image...

A Discriminative Approach to Robust Visual Place Recognition

“…These methods employ either regular cameras ( [5], [6]) or omni-directional sensors ( [7], [8], [9], [10], [11]) in order to acquire images. The main differences between the approaches relate to the way the scene is perceived, and thus the method used to extract characteristic features from the scene.…”

“…Local image features may also be regarded as natural landmarks. The SIFT descriptor [13] was successfully used by Se et al [14] and Andreasson et al [11] (with modifications), while Tamimi and Zell [6] employed Kernel PCA to extract features from local patches. Global features are also commonly used for place recognition.…”

Vision Based Reconstruction Multi-Clouds of Scale Invariant Feature TransformFeatures for Indoor Navigation

“…The focus of this study is how to make semantically environment representation for the blind person to navigate and recognize objects needed. Several other domains consider addressing this challenge; these include hierarchical representations of space, high-level feature extraction scene interpretation, the notion of a cognitive map and finally the field of Human Robot Interaction (HRI) [12] . The study presented here closely resembles those that suggest the notion of a hierarchical representation of space [11,12,21] .…”

“…The table referenced by a composite key of indexes for objects. The cloud of order two will include zero component instead of the missed object inside the cloud, the cloud of order two will has two zero component, this explained later on The clustering object into a cloud involves representing the blind person's environments as a large set of features, some features are very important to navigation process The construction of cloud with some determined objects features can be set as a landmark Landmarks are parts of the image which hold sufficient information about the image [12] . Usually small set of landmarks per images is needed.…”

Storing and recalling information for vision localization

“…Other local-feature such as Kernel PCA features [8] and Harris corners [9] are also used with varying degrees of success. One disadvantage of using local features is the number of features needed to be stored for each image.…”