•  
  •  
 

Journal of Innovation in Science, Engineering and Technology

Document Type

Original Study

Abstract

Color Vision Deficiency (CVD) is a medical condition that reduces the ability to distinguish certain colors and may occur due to genetic inheritance or acquired retinal damage. Drivers affected by CVD often experience difficulties at traffic light intersections because traffic signals rely entirely on color-based communication. Existing assistive systems mainly address this issue through audio feedback. However, audio-based solutions may perform poorly in noisy urban environments and are less effective for individuals with hearing impairments. The already existing assistive systems provide visual outputs in the dashboard that do not ensure all types of CVD drivers can see the visual output. This research addresses these gaps by developing an IoT-mobile assistive system to support drivers with CVD conditions. The system consists of three integrated components: an IoT device, a lightweight edge AI model, and a mobile application. The IoT device captures the traffic light signals, detects the signal using a lightweight edge AI model, and sends the detected signal to the mobile application. The mobile application receives the detected traffic light signal and displays it as a high-contrast visual output, which any type of CVD driver can identify. The AI model was initially trained using the YOLOv8s architecture and then converted to the TensorFlow Lite format. This lightweight AI model is deployed on the IoT device, which leads to edge processing that eliminates the reliance on an internet connection. The detected traffic signal is then transmitted to a mobile application via Bluetooth Low Energy (BLE) communication. This received signal is displayed in unique high-contrast symbolic indicators to identify the signal without any confusion. The mobile application provides audio aids for each signal, which helps to grab attention and identify the signal by audio in less noisy environments. The system prototype was evaluated under laboratory conditions to identify performance, accuracy, communication latency, and processing efficiency. The results demonstrate that deploying a lightweight edge AI model on a resource-constrained environment is feasible for real-time traffic light detection. Furthermore, the symbolic visual interface provides an accessible and practical assistive solution for drivers with CVD.

First Pages

88

Last Page

95

Download

Share

COinS