EFFICIENT OBJECT RECOGNITION USING HAAR CASCADES AND CONVOLUTIONAL NEURAL NETWORKS

Abstract

Surveillance, Security, and Healthcare require efficient Image recognition, which is critical for diverse applications requiring high accuracy and computational efficiency. This work combines HAAR Cascade classifiers with Deep Convolutional Neural Networks (CNNs) to present an efficient Object Recognition system that enhances detection accuracy while reducing computational overhead. Initially, the HAAR Cascade classifier rapidly detects regions of interest, minimizing the amount of data processed by the CNN. Subsequently, a CNN architecture, optimized through dropout regularization and Hyperparameter tuning, performs detailed classification on the detected regions. The system was evaluated using the MNIST and CIFAR-10 datasets, achieving 90% accuracy on CIFAR-10, 91% on a custom dataset, and maintaining 78% accuracy under conditions of occlusion and varying lighting, demonstrating robustness in real-world scenarios. This layered approach balances speed and precision, making it suitable for real-time deployment in surveillance, medical diagnostics, and automated security systems. The proposed system demonstrates that integrating traditional feature detection with deep learning techniques can yield a practical, scalable solution for advanced image recognition tasks.