Abstract:
Object detection remains one of the most researched areas in the field of Digital Image Processing. With the introduction of Convolutional Neural Network (CNN), there has been a revolution in the detection approaches. Although the detection algorithms have come a long way, detecting objects for the blind or visually impaired people (BVI) is a completely different sce- nario. Rather than the detection of objects, for the visually challenged people this task is primarily focused on obstacle detection. Based on this concept, several approaches have been made to design smart canes that can be used as a helpful walking tool. More robust approaches include real time imag- ing through camera devices and processing the images to detect objects or obstacles. It remains a challenge to ensure both sufficient performance and cost efficiency at the same time. In many cases, the design architecture is not convenient enough for the visually handicapped persons. Also very few attempts were made to combine depth information with an object detection method in real time.
In this paper, we propose a completely new system framework that per- forms detection for the visually challenged people. We use a depth sense camera and a portable computing device to analyze the depth data and combine with the detection method to detect objects and also obstacles in real time along with its relative position and also the distance from the user. We perform the object detection using YOLO (You Only Look Once) algo- rithm which is comparatively faster than almost any recent object detection algorithm. Even if an object is not detected by YOLO due to lack of light or any other cause, the depth information will allow us the detection of obstacle and also the position and distance can still be calculated. Finally the total information gathered in real time will be narrated with convenience to the subject.
