Staircase and escalator detection for visually impaired
AuthorZereen, Aniqua Nusrat
MetadataShow full item record
In this thesis, two different methods are presented for staircase detection. First method works with real time captured still image and second method with real time video. Stairway detection and identification of up stair and down stair is important for both independent and safe navigation of visually impaired. Proposed first method for detecting up stair and down stair using Gabor filter and Support Vector Machine (SVM). Energy distribution as a feature has been extracted from 40 filtered images using Gabor filters with 5 scales and 8 orientations. These features are trained and tested with four different classes, up stair, down stair, freeway and other. The overall classification accuracy of the proposed method is 92.9% based on experimental result. In second method, real time video is captured with Microsoft Kinect. Detection of real time moving object along with the moving direction in respect with visually impaired people is a challenging research area. The recent advancement in technology for real world scene capturing and portable devices like Microsoft Kinect necessitate the need of more robust and faster techniques for assisting blind navigation. The objective of this study is to develop a suitable and an effective technique for moving object detection along with its moving direction in indoor environment. Depth information of the font scene of a blind people is captured using Microsoft Kinect version 1. Three consecutive depth frames are extracted from video taken in one second and Distance Along Line Profile graph is generated for four predefined lines of each depth frame. These line profile graphs are analyzed for detecting presence of any moving object and the moving direction. After detailed investigation, experimental result shows that the proposed method can successfully detect moving object along with its direction and still objects with 92% and 87% accuracy respectively. The overall accuracy of the proposed second method is 90%.