From Pixels to People: Deep Learning Breakthroughs in Human Detection

The rapid progresses in deep learning (DL) and convolutional neural networks (CNNs) have revolutionised human recognition systems by enabling automated feature extraction and robust image analysis. This study evaluates the performance of Sequential Convolutional Neural Network (SCNN), VGG-16, VGG-19, ResNet, and Inception for human recognition tasks using the INRIA dataset. SCNN demonstrates a straightforward structure with impressive classification accuracy. VGG models leverage their depth for fine-grained feature extraction, ResNet handles vanishing gradients through residual connections, and Inception effectively captures multiscale features using its modular design. Training and validation results reveal that VGG-16 achieve remarkable performance with validation accuracies exceeding 97%, while ResNet presents moderate results, likely constrained by its complexity. The findings underscore the significance of tailored DL models for human recognition, offering theoretical insights into CNN architectures and practical implications for surveillance, autonomous systems, and human-computer interaction.