International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 05 | May 2024
www.irjet.net
p-ISSN: 2395-0072
CONVOLUTIONAL NEURAL NETWORK FOR HUMAN DETECTION Mr.S.VIGNESH KUMAR M.E.
SHEIK RASHATH D
Assistant Professor Department of Electronics and Communication Engineering Chennai Institute of Technology (Autonomous)
Department of Electronics and Communication Engineering Chennai Institute of Technology (Autonomous)
SAINARESH S Department of Electronics and Communication Engineering Chennai Institute of Technology (Autonomous)
----------------------------------------------------------------------------------***-----------------------------------------------------------------------------ABSTRACT -This paper investigates YOLOv8, a deep to detect certain objects inside an image, which can range learning model used for real-time human detection. Using from cars and humans to animals and household goods. its efficient architecture, YOLOv8 provides quick and However, identification is insufficient. Object detection precise object identification. Pre-trained models are demands localization, which identifies the actual locations utilized to perform initial human recognition inside a of these objects inside the image. This is commonly larger object classification. Frameworks such as PyTorch accomplished by creating bounding boxes around the and Ultralytics make implementation easier. Roboflow is observed items. By combining these two tasks, object used to train the model on a special human detection detection provides a more comprehensive grasp of the dataset with over 1000 annotated photos, potentially visual world within an image.YOLO (You Only Look Once) outperforming earlier methods in terms of accuracy. has been a game changer in object detection since its birth Preprocessing processes and model output (bounding in 2015, thanks to its efficient single-stage architecture. boxes with confidence ratings) are examined in the paper Unlike approaches that require numerous passes of an and methods for filtering detections and visualizing image, YOLO analyzes the entire image in one step, results are investigated.The paper also presents a method making it extremely rapid and well-suited for real-time for detecting humans using YOLOv8 which includes data applications. YOLO's progress through its different collection (human and human-like objects), annotation versions, from YOLOv1 to the most recent YOLOv8, has (labeling of objects), training, and optional fine-tuning. been characterized by an unwavering commitment to We describe YOLOv8's real-time applications such as video improving accuracy and efficiency. surveillance, emphasizing its potential for counting pedestriansand activity monitoring. Integration with YOLOv1, the first version, introduced the single-stage multi-object tracking methods is being investigated for detection technique, which achieved real-time rates but improved functionality.The study emphasizes the had lower accuracy than certain multi-stage detectors. In importance of fine-tuning pre-trained models using 2016, YOLOv2 achieved considerable improvements in human-centric datasets in increasing accuracy. It closes accuracy while preserving real-time performance. This by emphasizing YOLOv8's capabilities as a powerful and version included revolutionary approaches such as batch customizable solution for real-time human identification normalization and anchor boxes, which dramatically improved accuracy metrics. in a variety of circumstances. Keywords—Machine Learning, Deep Learning, Convolutional Neural Network (CNN),You Only Look Once(YOLO)
1. INTRODUCTION Object identification in computer vision extends beyond merely detecting image content. It aims to replicate human visual perception, allowing machines to not only detect but also precisely locate objects in an image. This goes beyond simply identifying "what" an image contains and tries to answer the queries "what" and "where" by recognizing items and their exact locations with bounding boxes. Object detection goes beyond basic image content understanding. It addresses two key goals by combining computer vision and machine learning approaches, frequently using deep learning models for improved accuracy. The first challenge is identifying things, similar to how humans see a scene. This requires teaching computers
© 2024, IRJET
|
Impact Factor value: 8.226
|
In 2018, YOLOv3 was updated with a more complicated architecture that achieved a commendable balance of speed and accuracy. This version was critical in establishing YOLO's status as a premier object identification technology. The third edition, YOLOv4 in 2020, represented another watershed moment by introducing new variations adapted to unique needs. YOLOv4-Tiny prioritized speed, catering to applications that required real-time performance, whereas YOLOv4-X valued accuracy, providing a solution for cases where precision was crucial. This adaptability enabled YOLO to meet a wider range of deployment requirements. Ultralytics introduced YOLOv5 in the same year, with the goal of simplifying the model and coding while keeping high accuracy. This iteration focused developer usability, reducing the deployment process and making YOLO more accessible to a larger audience.
ISO 9001:2008 Certified Journal
|
Page 2213