Research was conducted at West Virginia University under the sponsorship of the West Virginia Department of Highways to assess the feasibility of using machine vision to replace or augment manual inspection procedures in the construction of asphalt pavements.
A single camera was used to observe multiple aggregate particles arranged randomly on a light table.
The backlit image was thresholded to create a binary image. Mathematical image processing was carried out to separate touching and overlapping particles, find the particle edges and identify useful features of each particle.
The features included some fairly standard ones such as projected area, edge length, and centroid.
Several new features were also defined, based on an extension of Fourier spectrum techniques.
Combinations of the features were created to yield parameters that were translation-, rotation- and scale-invariant for use in a regression model.
The model was then calibrated and used to predict particle mass.
The results were encouraging, yielding errors of less than 2% for batch sizes of about 1900 particles in the range of 4.75 mm < particle sieve
size < 25 mm.