A New Improved Obstacle Detection Framework Using IDCT and CNN to Assist Visually Impaired Persons in an Outdoor Environment

In spite of the fact that various electronic assistive devices have been designed for individuals who are impaired visually in recent decades, but very few solutions have been developed for aiding them in order to recognize general objects in the environment. In any case, the studies in this area are gaining momentum. Among the different innovations being used for this reason, solutions based on computer vision are developing as one of the most encouraging alternatives principally because of their accessibility and affordability. This study presents an integrated approach towards detection of obstacles for assisting the visually impaired persons. Here, we employed Gaussian filtering based improved discrete cosine transform (IDCT) and convolutional neural network (CNN) (IDCT-CNN) which can enhance the detection accuracy. User defined functionality of DCT is implemented here as an advancement to existing DCT and results show better performance. Also, the proposed Gaussian filtering based IDCT-CNN performance is compared with the existing KNN classifier and related existing research. Results showed promising outcomes where the accuracy of the proposed approach outscored with 99% in comparison with the second best of morphological closing technique which achieved 81%.