A Deep Learning Framework for Virtual Drawing and Geometric Shape Prediction Using Convolutional Neural Networks

Hand gesture-based interaction provides a natural and effortless user experience to facilitate human-to-machine natural communication. This paper describes a deep learning approach for virtual drawing and geometric shape prediction based on computer vision and Convolutional Neural Networks (CNNs). The designed system combines real-time hand tracking, virtual canvas display, and a CNN-based classification model to identify user-drawn shapes through gestures. First, shapes on the virtual canvas are pre-processed—cropped, converted to grayscale, and resized-to provide consistent input for training. The CNN structure includes several convolutional and maxpooling layers, activated by ReLU and Softmax functions, to efficiently extract and classify shape features. The network is trained on a labeled training set of eight geometric shape classes, allowing it to learn subtle visual patterns. Training is accomplished by optimizing the network parameters via the backpropagation algorithm in order to reduce classification errors. Performance testing on an independent test set shows impressive results, where the model scores 98.37% accuracy. Important measures like precision and recall also support its efficacy. The learned model is integrated into a real-time virtual painting system, enabling users to draw and recognize freehand shapes without touch. The system enables applications in digital art, education, and gesture-based user interfaces, promoting humancomputer interaction. Future improvements can include multishape recognition, gesture-based color picking, and integration with augmented reality (AR) platforms.