Image deblurring for navigation systems of vision impaired people using sensor fusion data

Abstract

Image deblurring is a key component in vision based indoor/outdoor navigation systems; as blurring is one of the main causes of poor image quality. When images with poor quality are used for analysis, navigation errors are likely to be generated. For navigation systems, camera movement mainly causes blurring, as the camera is continuously moving by the body movement. This paper proposes a deblurring methodology that takes advantage of the fact that most smartphones are equipped with 3-axis accelerometers and gyroscopes. It uses data of the accelerometer and gyroscope to derive a motion vector calculated from the motion of the smartphone during the image-capturing period. A heuristic method, namely particle swarm optimization, is developed to determine the optimal motion vector, in order to deblur the captured image by reversing the effect of motion. Experimental results indicated that deblurring can be successfully performed using the optimal motion vector and that the deblurred images can be used as a readily approach to object and path identification in vision based navigation systems, especially for blind and vision impaired indoor/outdoor navigation. Also, the performance of proposed method is compared with the commonly used deblurring methods. Better results in term of image quality can be achieved. This experiment aims to identify issues in image quality including low light conditions, low quality images due to movement of the capture device and static and moving obstacles in front of the user in both indoor and outdoor environments. From this information, image-processing techniques to will be identified to assist in object and path edge detection necessary to create a guidance system for those with low vision.