Tutorials

Robots

Post Page Advertisement [Top]

artificial intelligencebookscomputer visiondeep learningimage processingmachine learning

Machine Learning and Robot Perception By Bruno Apolloni, Ashish Ghosh, Ferda Alpaslan, Srikanta Patnaik PDF

Download  Book: Machine Learning and Robot Perception By Bruno Apolloni, Ashish Ghosh, Ferda Alpaslan, Srikanta Patnaik PDF
Informations about the book:


Title: Machine Learning and Robot Perception

Author: Bruno Apolloni, Ashish Ghosh, Ferda Alpaslan, Srikanta Patnaik

Size: 25.8 MB

Format: PDF

Year: 2011

Pages:  357

Language: English



Country: India - Italy -Turkey - Australia

Book Contents:

This book presents some of the most recent research results in the area of machine learning and robot perception. The book contains eight chapters.
The first chapter describes a general-purpose deformable model based object detection system in which evolutionary algorithms are used for both object search and object learning. Although the proposed system can handle 3D objects, some particularizations have been made to reduce computational time for real applications.
The system is tested using real indoor and outdoor images. Field experiments have proven the robustness of the system for illumination conditions and perspective deformation of objects. The natural application environments of the system are predicted to be useful for big public and industrial buildings (factories, stores), and outdoor environments with well-defined landmarks such as streets and roads.

Fabrication of space-variant sensor and implementation of vision algorithms on space-variant images is a challenging issue as the spatial neighbourhood connectivity is complex. The lack of shape invariance under translation also complicates image understanding.The retino-cortical mapping models as well as the state-of-the-art of the space-variant sensors are reviewed to provide a better understanding of foveated vision systems in Chapter 2. It is argued that almost all the low level vision problems (i.e., shape from shading, optical flow, stereo disparity, corner detection, surface interpolation etc.) in the deterministic framework can be addressed using the techniques discussed in this chapter. The vision system must be able to determine where to point its high-resolution fovea. A proper mechanism is expected to enhance image understanding by strategically directing fovea to points which are most likely to yield
important information.
In Chapter 3 a discrete wavelet based model identification method has been proposed in order to solve the online learning problem. The method minimizes the least square residual parameter estimation in noisy environments. It offers significant advantages over the classical least square estimation methods as it does not need prior statistical knowledge of measurement of noises. This claim is supported by the experimental results on estimating the mass and length of a nonholonomic cart having a wide range of applications in complex and dynamic environments.

Chapter 4 proposes a reinforcement learning algorithm which allows a mobile robot to learn simple skills. The neural network architecture works with continuous input and output spaces, has a good resistance to forget previously learned actions and learns quickly. Nodes of the input layer are allocated dynamically. The proposed reinforcement learning algorithm has been tested on an autonomous mobile robot in order to learn simple skills showing good results. Finally the learnt simple skills are combined to successfully perform more complex skills called visual approaching and go to goal avoiding obstacles.

In Chapter 5 the authors present a simple but efficient approach to object tracking combining active contour framework and the opticalflow based motion estimation. Both curve evolution and polygon evolution models are utilized to carry out the tracking. No prior shape model assumptions on targets are made. They also did not make any assumption like static camera as is widely employed by other object tracking methods. A motion detection step can also be added to this framework for detecting the presence of multiple moving targets in the scene.

Chapter 6 presents the state-of-the-art for constructing geometrically and photometrically correct 3D models of real-world objects using range and intensity images. Various surface properties that cause difficulties in range data acquisition include specular surfaces, highly absorptive surfaces, translucent surfaces and transparent surfaces. A recently developed new range imaging method takes into account of the effects of mutual reflections, thus providing a way to construct accurate 3D models. The demand for constructing 3D models of various objects has been steadily growing and we can naturally predict that it will continue to grow in the future. 
Systems that visually track human motion fall into three basic categories: analysis-synthesis, recursive systems, and statistical methods including particle filtering and Bayesian networks. Each of these methods has its uses. In Chapter 7 the authors describe a computer vision system called DYNA that employs a threedimensional, physics-based model of the human body and a completely recursive architecture with no bottom-up processes. The system is complex but it illustrates how careful modeling can improve robustness and open the door to very subtle analysis of human motion. Not all interface systems require this level of subtlety, but the key elements of the DYNA architecture can be tuned to the application. Every level of processing in the DYNA framework takes advantage of the constraints implied by the embodiment of the observed human. Higher level processes take advantage of these constraints explicitly while lower level processes gain the advantage of the distilled body knowledge in the form of predicted probability densities.

Chapter 8 advocates the concept of user modelling which involves dialogue strategies. The proposed method allows dialogue strategies to be determined by maximizing mutual expectations of the pay-off matrix. The authors validated the proposed method using iterative prisoner's dilemma problem that is usually used for modelling social relationships based on reciprocal altruism. Their results suggest that in principle the proposed dialogue strategy should be implemented to achieve maximum mutual expectation and uncertainty reduction regarding pay-offs for others. We are grateful to the authors and the reviewers for their valuable contributions. We appreciate the assistance of Feng-Hsing Wang during the evolution phase of this book. 


A1

No comments:

Post a Comment

Bottom Ad [Post Page]