Dataset for Deep Learning Based Navigation Assistance for the Visually Impaired People

The Dataset was created by extracting features from the recorded video sequences using custom object detection model of YOLO. It has some generalized features for every frame, such as the total number of objects detected in a frame, object distribution according to the position of the frame, proximity of the objects for the 2 and 4-meter ranges, and movement tracking of all objects by counting how many objects are moving away or getting closer. For the detected object, there are some specific features such as object class, average and minimum depth values, distance estimation, a calculated angle from the center of the frame, and movement direction of the object, whether it is moving away or approaching the user. These features are very relevant for the dataset of the AI model that will predict the decision about the movement of visually impaired people. These features were stored in 2 types of formats: CSV Format and JSON Format. For the CSV format, the object-specific features stored for only the three nearest objects. For the JSON format, all the extracted features and the object-specific features for all the detected objects in a single frame have stored in a structural way.