Computer vision for rock aggregates

Sammanfattning: Measurement of size, shape and texture of aggregates (rockparticles) is often used in geological mapping, rockclassification, quarry and mine production. The size ofaggregates varies from micro-meters to meters, and the types ofaggregates may be classified as natural (i.e. sandstone, graveland 'natural blocks') and artificial (i.e. rock fragmentation).Methods of measurements may be based on manual, mechanical andcomputer vision (image analysis) approaches, of which thelatter has been applied for the last fifteen years. Theadvantages of computer vision method compared to the othermethods are: (1) high speed measurements with a potential highaccuracy; (2) the more tolerable limitation of aggregate size;(3) possible on-line systems for industrial application; and(4) measurements of objects which are manually inaccessible (i.e. blocks on a cliff ). in theresearch area of computervision, objects delineation (segmentation) has been achallenging subject since 1970's, especially for aggregateimages with random noise and overlapping of particles. Size,shape and texture of aggregates are also difficult to determinein applied geology, mining, mineral processing, quarryproduction, building industry, railway and high-wayconstructions.This thesis gives a summary of research achievements in thearea of computer vision for rock aggregates, and includes ninearticles for describing two computer vision systems (on-linesystems) for monitoring rock aggregates on a moving conveyorbelt and in a gravitational flow, respectively.The summary of the Ph.D. thesis, in addition to the contentsin the nine papers, presents also a literature review, and theseveral applications by using the computer vision systemsdeveloped by the author. The research achievements are based onthree sources of the published articles and reports. They are:(a) nine articles collected in this thesis, the major part inthe summary; (b) a Licentiate thesis by the author in 1994;some published articles and reports by the author or otherresearchers for evaluating and testing the systems or theoriesdeveloped or proposed by the author. The collected papers inthe thesis mainly deal with: (1) image acquisition foraggregates in motion (paper A); (2) an approach for estimationof average size of densely packed particles based on textureanalysis (papers (B); (3) automatic selection of aggregateimages for segmentation (Paper C); (4) an adaptive thresholdingalgorithm for image binarization (paper D); (5) an algorithmfor splitting touching particles in a binary image (paper E);(6) size and shape measurements, and comparison between sievingand image analyses (papers F-G); and (7) field study of on-linesystems for aggregates on a conveyor belt and in agravitational flow (papers H-I).Key words:image acquisition; image segmentation;splitting touching particles; aggregate particles; size, shapeand texture; gravitational flow; conveyor belt; rockpile; imagedistortion; gray value; edge density; split-and-merge.