RECOGNITION OF REALTIME BASED PRIMITIVE GEOMETRY OBJECTS USING PERCEPTRON NETWORK

Abstract

The purpose of this study is to analyze the perceptron model on pattern recognition of primitive geometric objects in real time based on video images. The samples used in this study were cubes, prisms, tubes and balls. The system was built using the Delphi 7 programming language with pre-processing stages system training includes the process of calculating matrix values from the original image, then proceed with the grayscale and edge detection processes using convolution with a kernel, namely the sobel operator and then the matrix results from the edge detection process are transformed using a perceptron network to obtain energy from the image of the object, then the resulting energy The transformation is stored in the database as a system test reference pattern recognition energy. Measurement of system performance evaluation in this study uses two parameters, namely detection rate and false positive rate. The recognition rate of primitive geometric objects using the perceptron network model in this study reaches 60.00% to 80.00%. The detection rate percentage shows that this model can be used as a supporting approach for the recognition of geometric objects in video.