Work

Spatial Sound Based Guidance System : Initial Study on the Usability of Spatial Audio and Object Detection Algorithms for an Assistive Guidance System

Year: 2024
Type: dissertation
Abstract: Over 43 million people worldwide are classified as fully blind.The loss of our most dominant sense has a big impact on a person's day-to-day life and also negatively affects those around us.This is partly the reason for developments in the field of electronic assistive devices.This thesis evaluates the usability of object detection and spatial audio for usage in microspace navigation task.The evaluation is mainly done through a proof-of-concept device that tells the user the approximate location of a target object using spatial audio simulation through headphones.The proposed system uses a modern object detection model (YOLOv11) and a depth camera with high accuracy (ZED 2i) to estimate the location of the target object in a 3D scene around the user.The spatial audio is simulated with a Python program with the help of third-party spatial audio software.The audio input is given to the user through headphones.The system is able to achieve near real-time processing speed allowing the synchrony between the user's and target object's movement and the audio input from the system.The system capabilities are tested through a testing scenario where voluntary participants are asked to find a target object from a table cluttered with objects belonging to different categories.The system is tested with two different settings, one where only the target audio is present and one where there is a reference audio source coming from the user's hand location.The hand location information is calculated by first detecting the hand using Google's MediaPipe hand detection model and then estimating the location using the depth map provided by the depth camera.With a low number of participants (N=4), the results are presented as only initial results.Further testing is needed to offer a more statistically significant discussion about the usability of object detection and spatial audio in microspace navigation task.The results from the initial study indicate that the system did offer some advantage in average times to find objects when the user had not explored the test environment beforehand.The more convincing results are that the system was able to reduce the number of non-target objects touched during the exploration for the target object by an average of 1 less object touched when using the system.The conclusion is that object detection and spatial audio showed promise in the domain of microspace navigation for the blind.Further testing is needed to prove the statistical significance of these results.
Language: English
Open Access status: green
OpenAlex ID: W7132918894
Cites: 39
Cited by:
Related to: 10