This work proposes a robust human action recognition and intrusion detection approach using computer vision and machine learning. It integrates two key components: a YOLOv8 model for detecting humans by identifying key points (such as shoulders and elbows) and bounding boxes, and a custom deep learning model for action recognition, which processes these key points to determine intrusions. The model analyzes the features in each video frame, classifying actions to determine if an intrusion occurs. Additionally, it uses Twilio to send SMS notifications when an intrusion is detected, providing real-time alerts. The architecture allows for easy integration of different deep learning models, making it adaptable to various security applications. This system offers a reliable solution for automated intrusion detection, combining deep learning with communication technology for real-time threat monitoring.