Tham khảo tài liệu 'innovations in robot mobility and control - srikanta patnaik et al (eds) part 14', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Using Visual Features for Building and Localizing within Topological Maps 253 ings to landmarks and that the landmarks may not be distinguishable. Other maximum-likelihood based methods such as Konolige 14 Folkesson and Christensen 7 and Lu and Milios 17 describe how to minimize an energy function when registering laser scan matches. Our work differs from this in that we linearize the energy function. While this simplification may not generate an optimal solution the method is not likely to be affected by local minima in the energy space during relaxation. Sim and Dudek 22 describe a visual localization and mapping algorithm which uses visual features to estimate the sensor readings from novel positions in the environment. In practice our vision system could be replaced by any other kind of boolean sensor modality which can report whether the robot has re-visited a location. In 29 a map is learned ahead of time by representing each image by its principal components extracted with Principal Component Analysis PCA . Brian Pinette 19 described an image-based homing scheme for navigating a robot using panoramic images. Krose et al. 15 and Artac 3 built a probabilistic model for appearance-based robot localization using features obtained by PCA. In 28 a series of images from an omnicamera is used to construct a topological map of an environment. Kuipers 16 learns to recognize places by clustering sensory images obtained with a laser range finder associating them with distinctive states disambiguating distinctive states in the topological map and learning a direct association from sensory data to distinctive states. A color signature of the environment is calculated using color histograms. Color information which is provided by most standard cameras is receiving increasing attention. Swain and Ballard 24 address the problem of identifying and locating an object represented by color histograms in an image. Cornelissen et al. 4 apply these methods to indoor robot .
