Prototyping Classification Device for Visually Impaired

Abstract

​

The project title is the prototyping of an object classification device utilizing computer vision to aid the visually impaired. The purpose of this experiment was to construct a portable prototype utilizing computer vision to allow the visually impaired to rapidly locate lost household objects. The hypothesis was that if an individual was equipped with the constructed prototype, they would find all the specified objects within the 10 minutes allotted and it would be faster than the control in which the individual did not have the device-equipped. In essence, the researcher constructed an object identification algorithm that utilized SIFT (scale-invariant feature transform) on a portable, Raspberry Pi computer that was also equipped with a camera. The control group was the trials in which the individual did not have the prototype, and the experimental group was the trails in which the individual did have the prototype. The researcher took note of how quickly each object was found and the misclassification rate in the experimental group. It was concluded that the prototype did not function as well as the control as it took nearly 3 more minutes to find all the objects with a relatively high error rate. Although, the prototype partially supported the hypothesis as all the objects were found within the 10 minutes. In the big picture, the findings of this research support a transition of novel artificial intelligence technologies to practical applications, such as those that help disabled people, at a low cost and high accessibility.