Identifying Buildings with Ramp Entrances Using Convolutional Neural Networks

The Americans with Disabilities Act (ADA) is a civil rights law that was signed into law in 1992 by President George H.W. Bush. The law requires wheelchair access be made available for buildings built after 1992. Buildings under the law include retail stores, hotels, banks and most other public buildings. However, there are a large percentage of buildings built before 1992 that are not wheelchair accessible. In addition, ADA does not require the location of ramp to be at the front of the building. This is an inconvenience for individuals who use wheelchairs to access a building, as a) the building may not have a ramp or b) they may have to roll around the building to where the ramp may be located. Hence, in this paper, we describe a prototype artificial intelligent system, which takes the input of a building image, and produces the output prediction for whether the building has a ramp. The system uses a deep learning technique, convolution neural network (CNN) to classify building images. We evaluated our method on a sample dataset of building images that we collected and building images from online sources. Training and validation accuracies were very high, 98.9 and 95.6 percentages respectively.