Navigation Assistant for Visually Impaired

Led a 4-person team to develop software on MATLAB using Computer Vision that generated a map from the surroundings that identifies objects in the environment and remembering past locations visited to generate a comprehensive map that enabled navigation with an accuracy of 40 cm

Developed an assistive navigation system based on simultaneous localization and mapping (SLAM) , to assist visually challenged people in both indoor and outdoor environments. The system uses a stereo camera and ORB to detect features in images and gives instructions to guide the user to the intended location by matching the detected landmarks against the corresponding features on the digitised floor map. Because of incorrect data association caused by moving objects, the accuracy of localization and mapping is reduced. Mapping accuracy utilising visual simultaneous localization and mapping (V-SLAM) is reduced in dynamic environment conditions. To address this issue, a stereo SLAM approach based on dynamic object rejection is developed, in which only the static portion of the image is extracted and used in mapping. This improved robustness as accuracy of the model in dynamic conditions.

This Navigation system included functionalities like object detection and recognition using YOLO, to help navigate around them.