An Research On Visual Image Processing Of Little Ground Mobile Robot

Deeper researchment has been made on mobile robot with visual system along with thecontinue development of computer technology, machine visual and image processingalgorithm. This paper has made researchment on the problems related to the visual imageprocessing of the mobile robot which works in the outdoor unstructured environment. Withthe expecting of raising the ability to percept and recognize the outdoor complex environmentof the mobile robot, researchment has been made on the visual image preproceesingalgorithm, feature extraction and classification.After deeper researchment on the image preprocessing being made, this paper introduedthe factor of the degree of blur (FDB) to evaluate the quality of the processed image.Comparison of the processed image quality has been made between histogram equalizationand specification, as well as the image quality filtered with the common used smoothingalgorithm.The contrast experiments show that when using histogram equalization to adjustcontrast and using meadian filter and gauss filter to wipe out noise get lower FDB.This paper has combined color, texture and shape as descriptor to better discrible animage. Color moment, gray level co-occurrence matrix (GLCM) and Hu invariant momentsare use to extract color, texture and shape respectively and some experiments have beenconducted to examine and certify the sensibility of these three algorithms to the rotated and(or) scale-changed image. The experiments show that these three algorithms are mostlyinvariant to rotated and (or) scale-changed image.For the common outdoor environments, five kinds of terrains (obstacles) including grass,road, trunk and stone are collected and their features are extracted. To raise the accuracy rateand reduce the cost of time, this paper has made qualitative and quantitative statisticalanalysis to the original features. Separable factor is defined to measure the separabilityamong these five terrains (obstacles) under particular dimension of feartures. It realizedfeature selection and dimensionality reduction by excluding features with lower separablefactor. Classification expertiments were conducted on the reducted dimensions of features,and results showed that the accuracy rate raised and time-cost reduced.At the end of this paper, the design of the software of terrains (obstacles) classification system is realized and proved by the outdoor terrains (obstacles) on the Voyager-â…¡ platformthat the functions are regular and the performance of the classification is favorable.