Real-time object detection is a crucial task in computer vision, enabling intelligent systems to identify and

a custom dataset enhanced through extensive data augmentation. This research demonstrates the potential of

integrating deep learning with web-based technologies for real-world applications such as surveillance, industrial

monitoring, and autonomous systems.

Convolutional Neural Networks (CNN).

In the modern era of Artificial Intelligence (AI), computer vision has become a vital branch that enables

computers to understand and interpret visual data. Object detection, one of the most fundamental applications of

computer vision, involves locating and classifying objects within images or videos. Traditional approaches such

as Haar Cascades and Histogram of Oriented Gradients (HOG) depend heavily on handcrafted features and lack

has potential applications in smart surveillance, autonomous driving, and accessibility enhancement for visually

impaired individuals [2].

The objectives of this project are centered around developing a robust, efficient, and scalable real-time object

detection system using deep learning techniques. The detailed objectives include:

 Accurate Detection: To design a YOLOv8-based model that can recognize and locate multiple objects in

real-time images and videos with high reliability, even in complex scenes.

 Speed Optimization: To ensure low-latency processing by optimizing model inference speed using GPU

acceleration and efficient model loading techniques for real-time performance.

 Scalability: To provide an easily extensible architecture where new object categories can be introduced

secure and consistent with ethical AI principles [3].

Research in real-time object detection has evolved significantly with the advancement of deep learning

frameworks. The study by R. K. et al. [1] presented a comparative analysis of YOLO-based architectures,

emphasizing how YOLOv8 improves upon earlier versions by adopting a decoupled head, enhanced backbone,

and anchor-free detection approach. This enables higher speed and accuracy for small and large objects alike.

Mohammed Kawser Jahan et al. [2] proposed an enhanced YOLOv8 model for improving online platform safety

by detecting objectionable content in live streams. Their work introduced fine-tuning techniques and

hyperparameter optimization that significantly reduced false positives. The integration of YOLOv8 with edge

rewarding models for successful detections.

Afdhal et al. [ 5 ] applied YOLOv8 for self-driving cars, emphasizing performance in mixed traffic environments.

Their research validated YOLOv8’s scalability and real-time efficiency, proving it to be an ideal choice for

critical applications like automated driving.

Traditional object detection methods are computationally expensive and inefficient in real- time scenarios. They

rely heavily on manual feature extraction and often fail in dynamic environments with changing light, occlusion,

or motion. Therefore, there is a pressing need for an automated deep learning–based system that can detect and

classify multiple objects in real-time from various input sources such as webcams, videos, and images, while

maintaining low latency and offering user accessibility through a web platform [1] [2].

Moreover, current detection solutions lack easy accessibility for non-technical users. They often require

installing specialized software or deep learning libraries locally. The absence of an integrated, browser-based

interface further limits their usability in practical applications such as surveillance or live analytics [4].

The proposed system integrates YOLOv8 with a Flask-based web platform to build a robust, efficient, and user-

friendly real-time object detection solution. YOLOv8 uses a CNN backbone for feature extraction, followed by

a detection head that predicts bounding

The architecture includes key modules: data acquisition (input from user), preprocessing (resizing,

normalization), model inference (YOLOv8 detection), and result visualization (bounding boxes, labels, and

scores). The integration of Flask enables a responsive, platform-independent environment that supports multiple

users and devices. Security features such as access control ensure safe use in restricted environments [4][5].

Real-time object detection will be done using a YOLOv8 model integrated with a Flask web interface.

It begins with the preparation of the data, including resizing, normalizing, and augmenting images to increase

model accuracy. The YOLOv8 model will be trained based on this dataset. Architecture:

Backbone: It extracts important features using CNN layers.

This interface can be used through Flask for uploading any files or using a live camera for detection. Results

processed will be instantly shown inside the browser fast, accurate, and user- friendly for applications like

surveillance and automation.

The following diagram represents the proposed system architecture for Real-Time Object Detection Using Deep

Learning. It illustrates the data flow from user input through the Flask web interface to preprocessing, YOLOv8

model inference, and output.

stream.

This study proposes a comprehensive real-time object detection system integrating YOLOv8 and Flask. The

system design prioritizes accessibility, scalability, and real-time interaction. Through a browser-based interface,

users can perform object detection on images, videos, or live feeds without technical expertise. This framework

lays a strong foundation for future extensions such as object tracking and smart surveillance applications[1][5]

1. R. K. et al., “A Perspective Study of Real-Time Object Detection Using Deep Learning,” IEEE