AI-BASED COMPUTER VISION SYSTEM FOR OBJECT DETECTION AND IMAGE RECOGNITION

Abstract

Computer vision, a critical branch of artificial intelligence, enables machines to interpret and understand visual data. This paper presents the design and implementation of a robust computer vision system for object detection and image recognition. The system leverages deep learning models, specifically Convolutional Neural Networks (CNNs) built with TensorFlow and Keras, and utilizes pre-trained architectures like ResNet and YOLO, fine-tuned on the COCO 2018 dataset. For feature extraction, we employ a convolutional layer-based approach to capture spatial hierarchies, complemented by Histogram of Oriented Gradients (HOG) for edge and texture information. Principal Component Analysis (PCA) is applied for feature selection to reduce dimensionality and computational overhead. The model was evaluated on performance metrics including accuracy, precision, sensitivity, specificity, and average recognition time. Results indicate a high-performance system with 93.4% accuracy, 91.2% precision, and an average recognition time of 140 milliseconds, demonstrating its potential for real-time applications in security, healthcare, and automation. This research contributes a scalable and efficient solution, addressing key challenges in object detection and recognition.

DOI: https://doi.org/10.5281/zenodo.19508447