The remainder of this page gives an outline and introduction to some of the industrial applications that have been tackled by various members of the group.
This research was aimed at performing an on-line measurement of the size distribution of rock material on the mill's feed conveyor belt and using that measurement to estimate the distribution of the contents of the mill. Image processing has been proposed as a solution for this application since it offers several advantages in the harsh environment of a mineral processing plant:
Results have been obtained with an average true detection rate of 69% and a further close miss rate of 14%, with very few false alarms (under various conditions of appearance and lighting). The overall result is that the measured distribution closely matches the true value for each of the test images.
Human inspection of the closures has proved to be unreliable and subjective, given the difficult working conditions (closures are manufactured at a rate of 20 per second by each machine). A machine vision system is capable of producing reliable and repeatable results in an environment which is unpleasant to humans. System development was divided into three sections:
Results have been obtained where 99.9% of the good closures are accepted and more than 90% of the defective closures are removed. These results have been obtained using a system installed on a production line.
Several such fruit sorting machines are available on the international market, at a cost of over R0,5 million. These machines are very successful in sorting certain types of dried fruit and particularly unsuccessful at others. A system was needed that:
Past research work in the field has resulted in a number of different techniques and algorithms for the purpose of segmentation. However, these algorithms are often iterative and therefore quite slow.The technique being developed is non-iterative and therefore one with industrial real-time processing implications.
A statistical comparison of the expected and observed distributions of bubble size, circularity and eccentricity revealed that the machine vision system does not mimic the human visual process exactly, but does however still achieve acurate segmentation which is favourable with regard to the expected results. In the final analysis, the computed parameters of bubble sizes and eccentricity can be used as robust and reliable measurements to describe the froth surface and bubble shapes. These parameters may thus be used to describe quantitatively the mineral grade and recovery values and be used in the optimsation of control of industrial flotation cells.
For further information please contact Professor Gerhard de Jager at
gdj@dip.ee.uct.ac.za