PAPER PRESENTATION ON

Multi Vision: Enhanced Object Segregation, Monitoring and Tracking using

YOLOv8 and DeepSORT

Authors:
Shrinidhi. S. S (142220104102)
Shriram. M (142220104103) Paper ID No: ICA5NT 396
Siva Prakash. S (142220104106) Project Guide:
Mrs. S. Shanthi, B.E, M.E.,
Assistant Professor
 Abstract
 This project focuses on developing a system for segregating, counting, and tracking objects
using YOLOv8 (You Only Look Once version 8) and DeepSORT (Deep Simple Online and
Realtime Tracking).
 Object segregation is crucial in various domains such as retail, surveillance, and traffic
management.
 YOLOv8 is employed for object detection, providing real-time and accurate bounding box
predictions.
 DeepSORT is then utilized for object tracking, enabling the association of detected objects across
frames to maintain their identities over time.
 The proposed system demonstrates promising results in accurately identifying, counting, and
tracking diverse objects, thereby facilitating better decision-making processes in various
applications.
 OBJECTIVE

 Implementing YOLOv8 for real-time and accurate object detection and focusing on achieving
precise bounding box predictions to facilitate effective object segregation.

 Utilize DeepSORT to associate detected objects across frames and enable the system to maintain
identities of objects over time, contributing to efficient tracking.

 Design the system to be scalable for handling varying complexities and scales of environments
and ensure adaptability to different scenarios by considering diverse object types and
environmental conditions.
 Existing System

 Object detection is often performed using methods like sliding window approach followed by
classification with classifiers such as Support Vector Machines (SVMs) or decision trees.
 YOLO is a deep learning-based object detection system that divides the input image into a grid
and predicts bounding boxes and class probabilities for each grid cell.
 It utilizes a single convolutional neural network (CNN) to simultaneously predict multiple
bounding boxes and their corresponding class probabilities.
 They often employ data association algorithms such as the Hungarian algorithm or deep learning-
based methods like DeepSORT (Deep Simple Online and Realtime Tracking).
 These systems leverage a combination of classical computer vision techniques and deep learning
algorithms to achieve object segregation, counting, and tracking capabilities.
 Proposed System
 The proposed system leverages the power of YOLOv8 and DeepSORT (Deep Simple Online and
Realtime Tracking) to address the shortcomings of existing systems in object segregation,
counting, and tracking.
 YOLOv8 is employed for real-time and accurate object detection, providing precise bounding box
predictions.
 DeepSORT integration facilitates the continuous tracking of detected objects across frames.
 The system is designed to address challenges in various domains, including retail, surveillance,
and traffic management.
 Beyond basic counting and tracking, the system contributes valuable insights for better decision-
making processes. Provides a deeper understanding of object dynamics in dynamic environments.
ARCHITECTURE DIAGRAM
 ALGORITHM OVERVIEW

 YOLOv8 divides the input into a grid and predicts bounding boxes and class
probabilities directly. Precise bounding box predictions and class probabilities for
detected objects are identified.

DeepSORT associates detected objects across frames using deep neural networks.
Maintains identities of objects over time, ensuring seamless and reliable object
tracking.

Overall, the combination of YOLOv8 and DeepSORT provides a powerful

framework for efficiently segregating, counting, and tracking objects in various
applications such as surveillance, traffic management, and retail analytics.
 ADVANTAGES OF THE PROPOSED SYSTEM
It leverages YOLOv8 for real-time and accurate object detection, ensuring swift
responses.

The advanced capabilities of YOLOv8 contribute to precise bounding box

predictions, enhancing the accuracy of object identification.

Integration with DeepSORT facilitates seamless object tracking, maintaining

identities across frames in dynamic environments.

Beyond basic counting and tracking, it provides valuable insights for better
decision-making processes.
 MODULES

Data Collection and Preprocessing

Object Detection

Object Tracking

Object Segregation and Monitoring

Integration and Fusion

1. Data Collection and Preprocessing:

 The Data Collection and Preprocessing module involves gathering video data from highway
surveillance cameras and other sources.

 cleaning and formatting the data to ensure consistency and compatibility, extracting individual
frames for analysis, annotating objects of interest for training purposes.

 This phase lays the groundwork for subsequent object detection and tracking algorithms by
providing clean, standardized, and annotated input data for accurate analysis of highway traffic.
2. Object Detection
 YOLOv8, renowned for its speed and accuracy in detecting objects within images or video frames.

 YOLOv8 divides the input image into a grid and predicts bounding boxes and class probabilities
for each grid cell simultaneously.

 YOLOv8's architecture, often based on deep convolutional neural networks (CNNs) like Darknet
or ResNet, allows it to handle various object sizes and classes efficiently.

 YOLOv8 can be fine-tuned on specific datasets to adapt to different environments or object

categories, making it a versatile choice for object detection tasks in dynamic highway
environments.
3. Object Tracking

 DeepSORT extends the traditional SORT (Simple Online and Realtime Tracking) algorithm by
incorporating deep learning-based features to improve tracking accuracy and robustness,
particularly in scenarios with occlusions and crowded scenes.

 DeepSORT associates detected objects across consecutive frames by leveraging both appearance
features and motion information, maintaining consistent track identities over time.

 By employing deep neural networks for feature extraction and similarity measurement,
DeepSORT effectively handles variations in object appearance and trajectory, enabling reliable
tracking of vehicles in complex highway environments with multiple lanes and diverse traffic
conditions.
4. Object Segregation and Monitoring

 The Object Segregation and Monitoring module focuses on categorizing and observing vehicles
detected and tracked in highway surveillance footage.

 The Object Segregation and Monitoring module focuses on categorizing and observing vehicles
detected and tracked in highway surveillance footage.

 This module segregates vehicles based on parameters such as lane position, speed, direction, and
behavior.

 It monitors various aspects of vehicle movements and interactions, including lane changes,
following distances, and adherence to speed limits.
5. Integration and Fusion

 The Integration and Fusion module consolidates the outputs of the object detection and tracking
algorithms, incorporating additional contextual information to refine object boundaries and reduce
false positives.

 Leveraging techniques like data fusion and contextual understanding, it refines the segmentation
and tracking of vehicles, enabling more precise monitoring and analysis of traffic patterns and
behaviors.

 Ultimately, this integration enhances highway safety and efficiency by providing comprehensive
and actionable insights into traffic dynamics and potential risks.
OUTPUT
 CONCLUSION

The proposed system, incorporating YOLOv8 for real-time object detection and
DeepSORT for seamless object tracking, presents a comprehensive solution for
efficient object segregation, counting, and tracking in dynamic environments.

The system's strengths lie in its real-time performance, accurate identification with
precise bounding box predictions, and adaptability to diverse domains.

The overall advantages make it a promising solution for applications in retail,

surveillance, and traffic management. The user-friendly design and scalability
further enhance its potential impact in diverse scenarios.