Falls, particularly among elderly individuals and patients with chronic conditions, pose a critical challenge in healthcare. Conventional monitoring techniques, including manual observation, bed alarms, and wearable sensors, often suffer from inaccuracies, false alarms, and patient non-compliance. These methods primarily respond to falls after they occur rather than preventing or predicting them. To overcome these challenges, this study proposes an intelligent monitoring system that leverages Computer Vision and Human Activity Recognition (HAR) techniques for real-time fall and vehicle crash detection. The system processes video streams by segmenting them into individual frames at 24 frames per second using OpenCV in Python. YOLOv8 is employed for object recognition, ensuring precise identification of people and vehicles. Fall detection is determined by analyzing bounding box dimensions—if the width surpasses the height, a fall is detected, triggering an alert. Similarly, vehicle crash detection is based on evaluating the proximity of detected vehicles; overlapping bounding boxes indicate a collision, prompting an alert. The system incorporates noise reduction and contour detection during pre-processing to enhance detection accuracy. Machine learning models trained on datasets such as COCO utilize feature extraction techniques to improve classification. Comparative analysis highlights YOLOv8’s superiority over Fast R-CNN and TensorFlow, achieving 77% accuracy for fall detection and 93.5% for vehicle crash detection. Real-time performance is optimized, and alerts are issued via email if anomalies persist for 20 consecutive frames. The system, deployed on a Flask-based web platform, demonstrates its effectiveness for healthcare, traffic monitoring, and emergency response applications.