DESIGN AND IMPLEMENTATION OF A REAL TIME OBJECT DETECTION SYSTEM USING MACHINE LEARNING

This project presented the design and implementation of a real-time object detection system using Python, developed to overcome the limitations of traditional approaches in processing complex and dynamic environments. The primary objective was to create an automated framework capable of accurately identifying, locating, and classifying multiple objects in both static images and live video streams with minimal latency. Such a system is relevant to a wide range of applications including surveillance, autonomous systems, healthcare, and smart environments, where precision, scalability, and speed are critical. The methodology centred on integrating the YOLOv8 model in ONNX format with a Flask-based web application to ensure accessibility through a browser interface. The development process involved several stages: dataset acquisition from open repositories such as Kaggle, preprocessing and annotation, model training and fine-tuning, and subsequent conversion of the trained model into ONNX for optimised inference. Deployment was achieved through a lightweight web interface capable of handling image uploads as well as real-time webcam feeds. Post-processing techniques were further applied to refine detections and enhance interpretability of results. Evaluation of the system demonstrated a balanced trade-off between detection accuracy and processing speed, enabling real-time responsiveness without compromising reliability. Comparative testing highlighted the model’s improved performance over conventional detection approaches in similar operating conditions. The study highlights the deep learning algorithms with efficient web technologies to deliver a scalable and user-friendly object detection solution. The project showcases the technical feasibility of real-time detection and provides a practical foundation for future research and deployment in diverse real-world applications.