Drowning remains one of the leading causes of accidental death worldwide, particularly affecting children and adolescents. In Peru, the growing popularity of both public and private swimming pools has increased the risk of aquatic incidents, especially due to the limitations of traditional supervision methods such as human lifeguards. This study proposes the design of an intelligent computer vision system for the automatic detection of people at risk of drowning in swimming pools. The system integrates classical descriptors (SIFT, HOG), machine learning algorithms (SVM, KNN, Random Forest), and state-of-the-art deep learning models (ResNet, ViT, PiT, DeiT, Swin Transformer, YOLO). A dataset of 3,712 labeled images was created through data augmentation, and several preprocessing steps were applied, including CLAHE, grayscale conversion, resizing, and normalization. Three methodological approaches were tested: (1) detection-only models, (2) detection followed by classification using machine learning, and (3) detection followed by classification using deep learning. The best overall performance was achieved by the Pyramid Vision Transformer (PiT) with hyperparameter tuning, reaching 98.38% accuracy, 99.23% recall, and an AUC of 99.77%. These results demonstrate the potential of the proposed system to enhance aquatic safety by reducing response times and supporting lifeguards in real-time detection of drowning incidents.