DefinePK

Abstract

Computer vision is the method of acquiring, processing, analyzing and understanding high-dimensional data images in order to produce numerical or symbolic information. These concepts are being pervasively applied in all fields of life, but due to the requirements of high processing power and high degree of parallelism, real-time computer vision applications are restricted to FPGA’s and high end processing GPU’s. The Raspberry Pi is a low cost medium processing power embedded development board. Raspberry’s capability in performing computer vision functions has been demonstrated with OpenCV support. But optimization of OpenCV functions in Raspberry Pi still considered a challenge. There is no direct OpenCV function for performing Connected component Labelling. The Connected Component Labeling is commonly used for identifying objects and marking fields for majority of computer vision application. This work discuses about the implementation and optimization of connected component labeling algorithms on Raspberry Pi. Apart from algorithm level adaptations for better hardware utilization, code level optimization is also explored. CCL is implemented with fastest algorithm known as LSL. In addition to LSL, Rosenfeld and contour based labeling are discussed for reference. The final implementation consists of real time connected component labeling with component numbering and estimation of area bounded by each component.