Analyzing the 3D Image Characteristics for a Pile of Hex Nuts using the Depth from Focus Method

This paper proposes the depth from focus method to analyze the 3D image characteristics in order to sorting hex nuts from piles. A fixed-focus lens with moving displacements produces multiple images with varying depths of focal field. 3D image characteristics of hex nuts are extracted as center coordinates, sizes, and tilt angles. Multiple nut images are convolved using a band-pass filter to evaluate the highest score from different displacements, and the three-dimensional height can be obtained by the depth gray-scale images. The individual nut regions can be discriminated by gray difference values, meaning less values classified as the same region. To improve the characteristic qualities, the region growing algorithm is used to segment the nuts and remove noise according to set area size groups. Comparing pixel coordinate and real-world coordinate leads to the tolerance of center coordinates, sizes and tilt angles for the precision analysis.