HUMAN BEING DETECTION SYSTEM USING CONVOLUTION NEURAL NETWORK

This project aims to develop a Convolutional Neural Network (CNN) system for the detecting the presence of human beings in images. Colour images were used and the images resized to a uniform size (227x227x3 pixels). Morphological image processing operations of reflection, translation, and saturation were applied to each image. Data augmentation technique was applied to avoid overfitting and to generalize well. The images were partitioned into Training Set (60%), Validation Set (10%) and Test Set (30%). The CNN model was trained with the images being labelled to direct the model to learn human presence (1) versus human absence (0). The system was trained for 5 epochs. Evaluation on a set of tested images provided qualitative insights, revealing generally accurate detection under favorable conditions. An overall accuracy of 96.22% was achieved, which demonstrates its effectiveness in classifying images containing humans versus those that do not. Notably, the system attained a perfect precision of 100%, indicating a strong ability to avoid false positive detections. A high recall of 92.98% further highlights the system's capability to correctly identify the majority of human instances. The balanced performance is reflected in an F1-score of 96.36%. Training curves indicated efficient learning and good generalization. The findings suggest a robust foundation for human detection, with identified areas for future work focused on enhancing performance in complex conditions and further reducing both false positive and false negative rates to improve overall system reliability and applicability.