Euro Banknote Recognition System for Blind People

Larisa Dunai Dunai, Mónica Chillarón Pérez, Guillermo Peris-Fajarnés, Ismael Lengua Lengua
2017 Sensors  
This paper presents the development of a portable system with the aim of allowing blind people to detect and recognize Euro banknotes. The developed device is based on a Raspberry Pi electronic instrument and a Raspberry Pi camera, Pi NoIR (No Infrared filter) dotted with additional infrared light, which is embedded into a pair of sunglasses that permit blind and visually impaired people to independently handle Euro banknotes, especially when receiving their cash back when shopping. The
more » ... detection is based on the modified Viola and Jones algorithms, while the banknote value recognition relies on the Speed Up Robust Features (SURF) technique. The accuracies of banknote detection and banknote value recognition are 84% and 97.5%, respectively. color. It is noted that the Euro banknote designs were developed in collaboration with the European Blind Union. In 1997, Klatzky and Lederman demonstrated that touch is one of the more efficient methods of object detection [6] . Currently, in order to be able to identify the banknotes, many blind users distribute them in advance, by value, into different pockets; this allows them to know the amount they are carrying. However, this classification requires qualifying time or a third person to help them. The artificial intelligence improvements and development of technologies have enabled great advances in the use of artificial vision for the recognition of the value of several currencies, such as Euro banknotes [7] [8] [9] , Dollars [10-13], Rupees [14], Mexican banknotes [15], the currency of Saudi Arabia [16], etc. Most of the work in banknote recognition is based on neural networks [15], Markov models, Principal Component Analysis (PCA) [13], or a Speed Up Robust Features (SURF) model [17]. This paper describes a system for the detection and recognition of Euro banknotes based on the application of Haar techniques proposed by Viola and Jones [18] and SURF [18] . The Haar features have been used for the detection of banknotes, and the SURF technique when identifying the banknote value. The Haar features are employed in order to identify the zone of interest in the image, instead of analyzing each pixel. This method allows for a drastically reduced computational time. By using Haar features, a set of local features is extracted afterwards, which are classified with the AdaBoost algorithm. The aim of the use of AdaBoost algorithms is to distinguish the Euro banknotes from complex images. The AdaBoost algorithm assigns a weight to each sample, and selects the feature that best classifies the sample according to the weight. Once the banknote is detected, the Speed-Up Robust Features algorithms are used in order to detect interest points in an image, each with their own characteristics. SURF algorithms use integral images, as well as the algorithms used on the banknote detection, which drastically reduces the computational time. The points of interest are detected by using the Fast-Hessian matrix. It describes the intensity content within the point of interest compared with neighboring items. Once the information of interest points and the neighboring items are recognized, the SURF descriptor is extracted from the region. Finally, features are matched between the trained image and the image acquired by the system. The paper is organized as follows: Section 2 enumerates the materials used on the prototype development. Section 3 describes the method of classification and detection based on Haar feature extraction, and the method of banknote recognition based on SURF methodology is also described. Section 4 summarizes the experiments and the results. Section 5 presents the conclusions of the work.
doi:10.3390/s17010184 pmid:28117703 pmcid:PMC5298757 fatcat:ulnq2ivyfvfmxom2cdce2mgvqm