Optimized Real-Time Object Detection and Counting in Video Streams With Yolov11 and Ssd Models

The numerous possible uses of crowd surveillance are attracting the attention of researchers. Examples of these applications could be tracking the reasons behind record preservation, identifying anomalous activity for security, and doing inventory for facility planning and expansion. This paper presents a real-time object detection and counting system that leverages the YOLO (You Only Look Once) and SSD (Single Shot MultiBox Detector) model for processing video files. The system is designed to automate object counting within a user-defined Region of Interest (ROI) in videos, offering applications in fields such as traffic monitoring, surveillance, and inventory management. The models were also compared with each other for the performance and it was found that YOLOv11 outperforms when compared to SSD model by 69.77 %. The user interface is built to have better clarity of the model performance, it is built with Gradio which simplifies the process by enabling non-technical users to upload videos, process them, and download the annotated results. This method is a powerful tool for item counting and real-time video analysis because it blends state-of-the-art machine learning algorithms with an easy-to-use front-end. It is shown that the suggested technique is accurate, quick, and useful for a variety of sectors.