Naimul Mefraz Khan
Software Systems
Advances in computer vision, machine learning, and depth interception has made it possible to provide a "virtual eye" to a visually impaired person. Through continuous analysis of an image captured through an RGB and depth camera, the "virtual eye" can provide spoken feedback on anything that is in front of it.
The objective of this project is to build a wearable computer (ideally head-mounted like a VR headset) that can provide automated analysis of objects and obstacles in front of a person while they navigate their surroundings.
1. The wearable computer should have a depth camera, an RGB camera, and a speaker (to play back audio feedback).
2. Images captured through the depth and RGB camera should be analyzed through a machine learning algorithm for object detection and localization (e.g. how far is the table from me?)
3. The feedback should be played back through the speaker in real-time to help the person in navigation.
1. Raspberry Pi/NVIDIA Jetson Nano with depth and RGB camera and speaker.
2. Analyze RGB image to detect objects through on-device machine learning (e.g. YOLOV4).
3. Map object bounding box to depth map to identify the distance of the object from the person.
4. Play analysis results back through speaker.
1. Survey hardware and software to figure out the best approach.
2. System design.
3. Implementation.
4. Testing under varying conditions (low lighting, fast motion, difficult angles, noisy environment etc.)
Hardware + wearable rig
Programming for image capture and audio playback
Programming for object detection through machine learning
Programming for mapping depth map to object detection output
NMK02: A Wearable navigation assistant for the visually impaired | Naimul Mefraz Khan | Monday August 30th 2021 at 12:44 PM