Obstacle Detection using Holoentropy-based Single Shot MultiBox Detector for aiding visually impaired persons

Abstract

People suffering from vision loss and sightlessness, face number of challenges, such as independent mobility and finding obstacles. Specifically, evading obstacles is a complex task for blind people because they greatly depend on their sense of touch and hearing. In such instance, they can observe the impediments at short distance only and they are failed to identify the obstacles, which are hard to determine. Hence, there arises a need for developing obstacle detection techniques to enable easy navigation of the visually impaired persons. Hence, this research work devises an efficient obstacle detection technique using Holoentropy-based Single Shot MultiBox Detector (HeSSD) for aiding visually impaired persons. Here, the input videos are extracted into various frames, and the obstacles are recognized and detected by using developed HeSSD method, where the matching process is carried by Holoentropy approach. Furthermore, the experimental result establishes that proposed HeSSDmethod attained highest accuracy of 0.80, MAP of 0.77 and recall values of 0.78.