Analisi Perbandingan Metode Perceptron Dengan Pseudo Invers Dalam Menentukan Matriks Fundamental Pada Citra Stereo Kamera

Abstract

3D object reconstruction is a technology that continues to develop and plays an important role in various industries. This technology allows the creation of three-dimensional models of real objects, which is useful in a variety of applications such as medicine, manufacturing, archeology, and entertainment. A commonly used method for reconstructing 3D objects is to use a stereo camera. This technology imitates the way the human eye sees the world by taking images from two different viewpoints and then combining them into a 3D model. This stage begins by taking 2D images from 2 cameras of the same type with different camera positions. Then, the 2D image is processed to find world coordinates and pixel coordinates with the center point being the center point. After that, the coordinate point data is processed to find the fundamental matrix value. The fundamental matrix is ​​a matrix that shows the geometric relationship between two cameras with different viewpoints in capturing 2D images. The matrix states whether the two cameras are calibrated or not. The matrix is ​​calculated using the pseudo-inverse method and perceptron artificial neural network (ANN). From these two methods, the mean squared error (MSE) for the pseudo-inverse method was 4.11 and for the artificial neural network method, it was 4.77. This shows that the pseudo-inverse method is better and more accurate than the perceptron artificial neural network (ANN) method for finding fundamental matrix values.