Performance Analysis of Object DetectionAlgorithms Using Standard Evaluation Metrics

Abstract

Object Detection is an integral and fundamental part of Computer Vision that helps in identifying and locating objects within images or video streams. It enables systems to focus on and learn from visual data effectively. Unlike image classification, which assigns a single label to an entire image, object detection provides both the class of objects and their spatial positions using bounding boxes and confidence scores. It combines the capabilities of image classification and object localization, allowing the detection of multiple objects simultaneously. It determines the class, type, and exact location of objects within an image. These capabilities make object detection highly effective and essential for numerous day-to-day and real-world applications such as surveillance, autonomous vehicles, traffic monitoring, healthcare imaging, and automated inspection systems. The output of an object detection model typically includes labeled bounding boxes, which indicate how precisely the objects are detected. These boxes accurately enclose objects of interest, including humans, animals, vehicles, and other entities. Designing, developing, and deploying models that ensure accuracy and computational efficiency in object detection remains a key research focus, as performance directly impacts reliability in practical applications. This work aims to study, evaluate, and prepare a dataset to analyze various object detection techniques. The evaluation is based on their ability to accurately classify and identify objects using standard performance metrics such as precision, F1-score, mAP (mean Average Precision), and inference time. The goal is to identify the most suitable approach for day-to-day, real-time, and real-world applications.