NAIN: Real-Time Object Detection and Navigation Assistance for the Visually Impaired using Single Shot Detector Model

Navigation for the Blind and Visually Impaired (BVI) poses a huge challenge due to lack of environmental cues. This research article proposes "NAIN" (Navigational Aid for the INsight-impaired) - an app for real-time object detection and navigation assistance for the BVI built using Python frameworks like Flask and OpenCV. Video frames from the camera are processed to recognize objects using the Single Shot Detector deep learning model implemented in TensorFlow. Detected objects are annotated with labels, proximity estimates, and communicated to the user via text-to-speech conversion and audio feedback. Object width in terms of pixels is mapped to real-world dimensions using the focal length of 500 pixels to provide distance estimates of objects like "Person detected at 300 cm". Quantitative testing shows over 90% accuracy in classifying daily objects. Distance approximation methodology provides 80% precise results for objects within 1000cm when properly calibrated. Initial trials illustrate reliable detection even in cluttered environments resulting in more confident indoor navigation. On the software architecture side, Flask enables graceful scaling of the detection pipeline across the client and server. HTML and CSS used in crafting an accessible interface optimized for use via screen readers. The system is designed for computational efficiency working in real-time without internet connectivity. It provides a template to develop affordable assisted perception for the BVI using computer vision. Future efforts look to enrich object libraries, integrate mapping APIs for outdoor settings, and provide navigation vectors around obstacles. By converting visual data into meaningful and intuitive audio cues, to raise awareness that can empower the BVI with greater mobility and independence.