International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056 p-ISSN: 2395-0072
Blind Assistance Using AI with Machine Learning Keshava Gowtham N.S1, S Harsha2, Shiva Subramaniyam S3, Shreesha G Hegde4, Rishi B.V5 1Assistant Professor, CIIRC Jyothy Institute of Technology Bangalore, India 2Dept. Of C.S.E, Jyothy Institute of Technology Bangalore, India 3Dept. Of C.S.E, Jyothy Institute of Technology Bangalore, India 4Dept. Of C.S.E, Jyothy Institute of Technology Bangalore, India 5Dept. Of C.S.E, Jyothy Institute of Technology Bangalore, India
---------------------------------------------------------------------***--------------------------------------------------------------------Abstract - Visually impaired individuals face numerous
challenges in navigating their environment independently. This paper proposes an AI-based blind assistance system that combines computer vision techniques, object detection using Optical Character Recognition (OCR), and gender detection using Convolutional Neural Networks (CNN) on a Raspberry Pi platform. The system is designed to provide real- time audio feedback for object recognition, text reading, and gender identification, enabling enhanced situational awareness. The proposed system aims to be a cost-effective and portable solution to improve the quality of life for visually impaired users.
Key Words -Blind assistance, machine learning, artificial intelligence, object detection, speech synthesis.
1. INTRODUCTION Artificial intelligence (AI), particularly machine learning (ML) and deep learning, has become pivotal in image processing and computer vision. Object recognition, a task humans perform effortlessly, remains a significant challenge for automated systems. This capability is crucial for developing applications that enhance human life, such as assistive technologies for the visually impaired and systems for improving road safety. Visually impaired individuals face numerous difficulties in daily life, from navigation to interpreting social cues like facial expressions. Technology can bridge this gap by providing tools that interpret visual information, such as detecting objects, recognizing faces and their emotions, identifying gender, and reading text aloud. Similarly, road safety is a major concern globally, with accidents often caused by undetected obstacles, poor visibility, or driver inattention. Systems capable of detecting obstacles like other vehicles, pedestrians, animals, or road hazards in real-time can significantly reduce accidents and improve driver awareness. This paper reviews and synthesizes approaches from five studies focusing on leveraging ML and deep learning for gender classification and object detection. These studies explore various algorithms, including SVM, CNNs implemented on platforms like Raspberry
Impact Factor value: 8.315
Pi, to create practical solutions for visually impaired assistance and accident prevention. The goal is to present a consolidated view of the methodologies, results, and challenges in these domains.
2. LITERATURE REVIEW / RELATED WORK This section reviews significant studies on gender classification, object detection, and customer journey analytics using AI techniques, providing the foundational background for the proposed system 2.1 Human
Learning,
Gender
Classification
using
Machine
Mali and Patil [1] compared SVM, CNN, and LDA models for gender prediction using facial images, with SVM achieving the highest accuracy of 98%. Despite its effectiveness, the system struggled with occlusions and limited itself to binary classification. 2.2 AI-Based Assistance for the Visually Impaired,
Safnaz et al. [2] proposed a CNN-based system integrated with OCR to aid visually impaired individuals through auditory feedback. The model performed well in real-time applications, although its emotion recognition accuracy was limited and hardware reliance affected scalability. 2.3 Deep Learning for Object Recognition,
Erastus et al. [3] developed a CNN model using the CIFAR-10 datasets, achieving 98.5% accuracy with the Adam optimizer. While deep learning proved superior to traditional techniques, it faced challenges with overlapping objects and high computational costs. 2.4 Object Detection for Assistive Technology,
Muhsin et al. [4] implemented a object detection system on a Raspberry Pi platform. The model provided real-time visual assistance but was constrained by object overlap issues and a fixed category set.
ISO 9001:2008 Certified Journal