USER PROFILE MANAGEMENT USING DEEP LEARNING FOR ASSISTIVE HEADGEAR
Siddharth Dhondiyal, Harshit Goyal, Shubham Ahuja, Nishtha Jatana · Proceedings on Engineering Sciences · 2025
Individuals with visual impairments face significant challenges in interacting with their environment, particularly in identifying and recognizing people. The learning disability associated with visual impairment compounds these difficulties, hindering their ability to navigate and engage with their surroundings effectively. This study proposes an innovative AI-powered user profiling solution using a multi-task cascaded convolutional neural networks (MTCNN) algorithm for rapid face detection, FaceNet for accurate face recognition, region-based and within-region priority setting, and a tailored feedback loop with a voice-assisted contact-saving feature. The research also aims to integrate the user profiling solution into an assistive headgear prototype called AXIE. Networking involves server and client-side communication via Message Queuing Telemetry Transport(MQTT) for maintaining low network latency. Experimental results confirm the successful integration of the user profiling system with the headgear. Our profiling model, trained using Labeled Faces in the Wild(LFW) and VGGFace2 datasets, demonstrated a detection accuracy(FD) of 99.23% and a face recognition accuracy of 95.3% (FR). Additionally, the network latency (NL) was measured at 60ms. Developing a user profiling system and integrating it with an assistive headgear marks a significant step towards empowering the visually impaired.