Title: Human Detection and Tracking

Image Processing

Project Details

Project Information

Project Title: Title: Human Detection and Tracking

Category: Image Processing

Semester: Fall 2025

Course: CS619

Complexity: Complex

Supervisor Details

Project Description

Title: Human Detection and Tracking

 

Project Domain / Category

 

Image Processing/ Research

 

Abstract / Introduction

 

Object detection and tracking is one of the rapidly growing areas of computer vision. We can use it in different domains. In this project, we are using it to analyze the data of CCTV footage from retail stores. We will analyze this data to detect humans and they’re in ouT count. After detection we will analyze data for buyer to non buyer ratio.

 

Functional Requirements:

 

        Data collection which is CCTV footage from retail stores.

 

        The system shall optimize the video for processing by resizing frames to a standard resolution (e.g., 640x480).

 

        The system shall perform frame enhancement if needed (e.g., brightness/contrast adjustment).

 

        The system shall use an object detection algorithm (e.g., YOLO, SSD, or OpenCV DNN) to detect humans in each frame.

 

        Human count: The system shall assign a unique ID to each detected person using tracking algorithms.

 

        The system shall track movement of individuals frame-by-frame.

 

        The system shall allow the user to define a virtual entrance/exit line within the video frame.

 

        The system shall detect entry (IN) when a person crosses the line into the store area and exit (OUT) when the person crosses in the opposite direction.

 

        The system shall keep track of cumulative IN and OUT counts for the selected video period.

 

        Analysis of data on the basis of in out and buyer non-buyer ratio in graphical manner

 

        GUI for users to interact with the system and to show graphical analysis of the system.

 

Non-Functional Requirements (NFRs)

 

        Performance Requirements

 

The system shall process video frames at a minimum rate of 10–15 frames per second (FPS) on standard hardware for real-time detection.

 

Detection accuracy for human presence should be at least 85–90% under normal lighting conditions.

 

The system shall maintain consistent tracking even in crowded scenes with partial occlusion.

 

Scalability

 

The system shall be designed to handle multiple video sources (e.g., more than one CCTV camera) without significant performance degradation.

 

The architecture should support future integration of more advanced detection models or cloud-based deployment.

 

 

 

 

 

 

Page 156 of 167

 

        Usability

 

The system interface shall be simple and intuitive, allowing users to easily select video files, set virtual lines, and view analytical results.

 

Visual results (counts, ratios, graphs) shall be displayed in a clear and understandable manner for non-technical users.

 

Reliability

 

The system shall be able to recover from temporary frame loss or video interruptions without requiring a restart.

 

Tracking IDs should remain consistent across frames to ensure accurate counting.

 

Security

 

Access to video data and analytical reports shall be restricted to authorized users. The system shall ensure that all stored footage and analysis results are protected from unauthorized access or modification.

 

Maintainability

 

The system code shall be modular to allow easy updates or algorithm replacements (e.g., switching from YOLOv5 to YOLOv8).

 

Documentation for setup, dependencies, and model training shall be provided for maintenance and further development.

 

Portability

 

The system shall be compatible with common operating systems such as Windows and Linux. The solution should be deployable on both local machines and cloud environments.

 

        Accuracy & Robustness

 

The system shall maintain detection reliability under varying lighting and crowd conditions. False positive and false negative rates should be minimized through threshold tuning and model optimization.

 

        Data Integrity

 

The system shall ensure that detected counts (IN/OUT) and analytical results are stored accurately without duplication or loss.

 

        Visualization Quality

 

Graphical analysis (buyer-to-non-buyer ratio, hourly traffic trends) shall be dynamically updated and exportable as images or PDF reports

 

Tools:

OpenCV

 

Image AI

 

Tensor flow

YOLOV3, Keras

 

Supervisor:

Name: Hina Rafique

 

Email ID: hina.rafique@vu.edu.pk

 

MS Teams ID: hina.rafique@outlook.com

Languages

  • Python Language

Tools

  • OpenCV, ImageAI, TensorFlow, YOLOv3, Keras Tool

Project Schedules

No schedules available for this project.

Viva Review Submission

Review Information
Supervisor Behavior

Student Viva Reviews

No reviews available for this project.