Year: 2024
Type: article
Abstract: This research explores the development of a multi-model obstacle detection and navigation system utilizing deep learning techniques to enhance the mobility of visually impaired people. The proposed system integrates various deep learning architectures, including a modified SSD Mobile Net, to achieve real-time obstacle detection and distance estimation. By employing a dataset comprising both indoor and outdoor environments, the system leverages neural architecture search to optimize the object detection framework, ensuring efficient processing on embedded devices. A key innovation of this approach is the incorporation of multi-sensor data, which enhances the robustness and accuracy of obstacle detection. The system utilizes advanced convolutional neural networks to process inputs from various sensors, including time-of-flight sensors, enabling it to identify obstacles with high precision and providing audio to user. The performance metrics indicate that the model achieves a mean average precision exceeding 90%, demonstrating significant improvements in detection speed and accuracy compared to traditional methods.
Authors: Jothilakshmi M, Praveen Kumar F, Surya Kumar A, Tamilselvan, Vignesh G
Institution: Sengunthar Engineering College (Autonomous)
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4405580293