Double Input Convolutional Neural Network Model for Determining Heights of Figures in 2D Images from Their Shadows

Abstract

This research developed a model using convolutional neural network (CNN) architecture for pattern detection and feature extraction from shadows and shapes in images. The dataset was constructed with photographic images of figures, with their respective shadows cast from different angles and locations. A DataFrame was constructed that combined with images of the figures improves the accuracy of the algorithm. Therefore, the presented model offers a favorable structure for estimating the height of 2D figures from shadows in images. The convolutional design makes it suitable for learning complex patterns in visual data, while the linear activation function in the output layer enables regression tasks. Thus, the optimized CNN model represents an automated process that learns to map patterns in shadow images to corresponding heights.

Keywords

• shadows
• shapes
• convolutional neural network
• dataset
• dataframe
• 2d
• height