Evaluation of an Integer Optimized Shape Matching Algorithm
Gernot Fiala, Johannes Loinig, Christian Steger · 2021
Computer vision and machine learning algorithms are often used for quality control for industrial products. Nowadays, neural networks can perform very well to detect the desired objects. Sometimes, the system has limited resources and is not capable of processing complex algorithms or use neural networks. Here, simpler algorithms are used for shape or object detection. The scope of the present work is to even lower the complexity of the shape matching algorithm by converting a shape detection algorithm to an integer version and evaluate the results. This allows to remove floating-point units (FPU) of processors and reduce the area of a System-on-Chip (SoC) design of a smart image sensor.