The Development of Mobile Application for Object Recognition Based on Deep Learning to Assist People with Visually Impaired

People with visual impairments often rely on tools like a white cane to assist them in their daily activities. They also use their sense of touch, hearing, and smell to navigate their surroundings. However, these tools and senses have limitations. For example, the sense of touch is ineffective for objects at a distance, while hearing and smell cannot identify all types of objects. Navigating new environments, such as unfamiliar outdoor areas, can be challenging and even dangerous for individuals with visual impairments due to the potential for collisions with moving objects. To address these challenges, our research aims to develop a mobile application that utilizes object recognition technology to assist people with visual impairments. The object recognition model is built using the SSD-MobileNet-vl architecture. Our experimental results demonstrate that the model can achieve a confidence level of more than 60% in real-time object detection. Additionally, the model's performance is influenced by the lighting conditions, with brighter conditions yielding better results than darker conditions for detecting moving objects.