EyeForBlind 👁️

👀 Overview

EyeForBlind is a revolutionary mobile application developed to visual impaired individuals, helping them navigate safely and independently in complex urban environments.

Key Features:

• Critical Object Detection
  • Identifying pedestrians
  • Detecting traffic signals
  • Locating pedestrian crosswalks
• Traffic Signal Analysis
  • Interpreting traffic light states (green, red, yellow)
  • Recognizing pedestrian crossing signals
• Safety Behavior Prediction
  • Tracking pedestrian movements
  • Predicting movement intentions
  • Alerting potential collision risks
• User-Friendly Communication
  • Converting visual information to audio
  • Providing timely and clear warnings

🎥 Demo

🚀 Quick Start

Prerequisites

• Python 3.7.0
• PyTorch
• OpenCV
• CUDA (optional, for GPU acceleration)

Setup Steps

1. Clone the Repository

   git clone https://github.com/HungDongne/Eye-for-blind.git
   cd Eye-for-blind

2. Create Virtual Environment

   python -m venv .venv
   .venv\Scripts\activate  # On Windows
   # source .venv/bin/activate  # On macOS/Linux

3. Install Dependencies

   pip install -r requirements.txt

4. Download Trained Model

   • Download the pre-trained YOLOR model from Google Drive
   • Move yolor_p6.pt to the yolormodel folder

🖥️ Basic Usage

# Run with default settings
python superman.py

# Specify video and enable GPU
python superman.py --videoName demo2.mp4 --use_cuda 0

Command Line Arguments

General Arguments

Argument	Type	Default	Description
--use_cuda	int	0	Enable GPU processing (0: CPU, 1: GPU)
--videoName	str	'demo2.mp4'	Input video file name
--resize	bool	False	Resize input video frames
--size	int	1280	Resize dimensions
--video2frames	bool	True	Convert video to individual frames

Object Detection Arguments

Argument	Type	Default	Description
--device	str	'cpu'	Processing device (cpu or cuda)
--weights	str	'yolormodel/yolor_p6.pt'	Path to pre-trained model weights
--img-size	int	640	Inference image size (pixels)
--names	str	'yolormodel/data/coco.names'	Path to class names file
--cfg	str	'yolormodel/yolor_p6.cfg'	Model configuration file path
--conf-thres	float	0.25	Object confidence threshold
--iou-thres	float	0.5	IoU threshold for Non-Maximum Suppression
--classes	int[]	None	Filter detection by specific class IDs
--agnostic-nms	flag	False	Enable class-agnostic Non-Maximum Suppression
--augment	flag	False	Enable augmented inference

📞 Contact

• Dong Quang Duy Hung - hungdqd1711@gmail.com
• Le Quy Duong - lequyduong1822@gmail.com
• Vo Le Hieu - huay1602@gmail.com
• Ha Truong Giang - hatruonggiang222@gmail.com