An insight into assistive technology for the visually impaired and blind people: state-of-the-art and future trends

Abstract

Assistive technology for the visually impaired and blind people is a research field that is gaining increasing prominence owing to an explosion of new interest in it from disparate disciplines. The field has a very relevant social impact on our ever-increasing aging and blind populations. While many excellent state-of-the-art accounts have been written till date, all of them are subjective in nature. We performed an objective statistical survey across the various sub-disciplines in the field and applied information analysis and network-theory techniques to answer several key questions relevant to the field. To analyze the field we compiled an extensive database of scientific research publications over the last two decades. We inferred interesting patterns and statistics concerning the main research areas and underlying themes, identified leading journals and conferences, captured growth patterns of the research field; identified active research communities and present our interpretation of trends in the field for the near future. Our results reveal that there has been a sustained growth in this field; from less than 50 publications per year in the mid 1990s to close to 400 scientific publications per year in 2014. Assistive Technology for persons with visually impairments is expected to grow at a swift pace and impact the lives of individuals and the elderly in ways not previously possible.

Appendices

Appendix A: Database construction

Four major scientific databases—IEEE Xplore database ¹², Thomson Reuters Web of Science ¹³, ACM Digital Library ¹⁴, and Elsevier ScienceDirect ¹⁵ were used as sources of quality multidisciplinary publications related to research in assistive technology for the visually impaired and blind people published in journals, conferences and book chapters. The bar chart in Fig. 3 illustrates the proportions of publications captured from the different scientific database sources. While the search within Web of Science, IEEE Xplore, and ACM Digital Library was restricted to a topical search from multiple fields containing relevant key words or phrases, the search within ScienceDirect was restricted to computer science, engineering, medicine, and design. A range of keywords and phrases were used as search terms including—wearable assistive devices, electronic travel aids, portable assistive device, navigation systems, assistive technology solutions, rehabilitation technology, vision substitution systems, etc all in the context of the blind and visually impaired. The search engines in the scientific databases matches on each key word individually as well as on the whole phrase with the descriptive stored metadata for each published document, which includes the title, publication title, abstract and author-defined keywords. Many articles on just pure medicine, neuroscience, psychology, general disability and impairments did appear in later search listings and were not considered as there was no any clear relation to engineering or technology. These terms restricted the type of research publications that we consider to that published in engineering, rehabilitation engineering, assistive and rehabilitation technologies, computer vision, augmented reality, virtual reality, sensor processing and technologies, cognitive science, vision rehabilitation and sensory substitution research.

The information analysis was carried out using Xpath (XML Path Language) and XSLT 2.0 (EXtensible Stylesheet Language Transformations), both of which are best suited for navigating, manipulating, querying, or transforming of XML data. It was first necessary to convert the results of publication search to a readable document in a non-proprietary format. This was achieved in two steps. First, the search results were exported to a personal library in Zotero (http://www.zotero.org), a powerful research tool to organize and analyze sources. Zotero comes with several styles for creating bibliographies. One can automatically export all of the references in a certain library in Zotero to specific bibliography formats like BibTeX or Endnote XML. XML is a markup language for representing structured information like documents, books, transactions, etc in a format that is both human-readable and machine-readable. XML markups are a form of document metadata.

A separate XML database was maintained for the search listings from each scientific database. Within XML, the metadata content concerning each publication is enclosed within standardized tags of elements. Hence, these were then accessible through XPath queries and expressions meant to match specific patterns and retrieve results. An XML editor provides the platform to perform XPath queries and regular expression searches and XSLT transformations meant for information analysis. Before the information analysis, it is essential to have the publication databases of metadata cleaned up of duplicate entries which arise from overlapping search terms. The database also needed to be preprocessed as it was found that unexpected characters, strings of symbols did creep into XML elements during the process of capturing the metadata from the source. The end goal is to have a well formed XML document for analysis.

Appendix B: Database analysis

The complete database with metadata content of publications was then analysed using a variety of methods. A survey of topics provides a broad introduction to the research area. Hence, the first step of analysis was to survey the publications by topic. As a useful starting point, we retrieved the specific fields i.e. titles and abstracts from the publication database through a pattern matching program in XSLT. Next we extracted common topics (based on their frequency of occurrence in the ‘titles’ and ‘abstracts’ of the publications) through a word frequency counter in XSLT and listed them. Unfortunately the most frequent words and collocations are found to be stop words. Stop words are extremely common words that are of little semantic value and so they must be excluded from analysis. Stop words like ‘the’, ‘and’, ‘of’, ‘have’ and non-informative words like ‘proposed’, ‘developed’, ‘approach’, ‘towards’, etc were compiled into a stop word list and filtered out. This left us with a refined list of most frequent words (Table 2 lists 100 most frequent words), that identifies the most interesting and informative words or topics in the field of assistive technology for the visually impaired and blind people today. The same corpus of title and abstract information from the publications database was used to generate a word cloud from Wordle¹⁶, a web-based tool. The static word cloud (Fig. 2) visualizes our analysis giving greater prominence to the more frequent words. Except for a mismatch of two words, the word list and rankings from our XSLT word frequency program matched that of Wordle’s program.

Our second step of analysis considered identifying the leading journals and conferences which publish or disseminate knowledge in the field of assistive technology for the visually impaired and blind people. These steps were performed with XPath queries (using path expressions) on the XML publication database. We gathered additional information, such as the proportion of published content by journals and conferences and year besides the reputation of these journals and conferences. The total number of journal articles, conference publications, book section publications are mentioned in Sect. 5. The analysis by journals and conferences are published in Tables 3 and 4 respectively; while the analysis by year is plotted in Fig. 4 and are discussed in detail in the main text of this article. This was followed by the adding up the number of publications each year in our database in order to determine the growth of journals and conferences in the field in the last two decades. The expansion of research interest and developments in this field is reflected in the increased number of publications within individual journals and conferences (Fig. 5).

Our next step was to consider the semantic environment of the most frequent words by analyzing their collocations. In corpus analysis, examination of word collocations is known to reveal the main underlying themes in the corpus. In this step, we employed a list (of most frequent words) longer than the one presented in Table 2. The goal was to obtain a fairly good overview of the main themes and sub-disciplines in the field of assistive technology for the visually impaired and blind people. Using absolute frequency as a collocational measure is the simplest method, but it is not recommended, as it does not lead to interesting results always. A more appropriate and widely accepted measure for statistical significance is the collocation metric based on Log-Likelihood Ratio. In collocation discovery the likelihood ratio examines the likelihoods $L(H_0)$ and $L(H_1)$ of two hypotheses $H_0$ and $H_1$ about words $w_1$ and $w_2$ [79]. If N is the total number of words in the corpus, and $c_1$, $c_2$, and $c_{12}$ denote the occurrences of $w_1$, $w_2$, and bigram $w_{1}w_{2}$; then maximum likelihood estimates for $p = c_2/N$, $p_1 = c_{12}/c_1$ and $p_2 = (c_2-c_{12})/(N-c_1)$. Assuming binomial distribution, the log of the likelihood ratio $\lambda$ is calculated as follows:

$$\begin{aligned} log \lambda= & {} log \frac{L(H_0)}{L(H_1)} \nonumber \\= & {} log \frac{b(c_{12}, c_1, p)b(c_2 - c_{12}, N - c_1, p)}{b(c_{12}, c_1, p_1)b(c_2 - c_{12}, N - c_1, p_2)} \nonumber \\= & {} log L(c_{12}, c_1, p) + log L(c_2 - c_{12}, N - c_1, p) \nonumber \\&- log L(c_{12}, c_1, p_1) - log L(c_2 - c_{12}, N - c_1, p_2)\nonumber \\ \end{aligned}$$

(1)

where L(k, n, x) = $x^{k}(1-x)^{n-k}$. On applying the log-likelihood test to our database consisting of preprocessed titles and abstracts from 3010 publications on the concerned field of research, we obtained the most interesting two-word collocations and ranked them based on their corresponding LLR scores (Table 5 lists 50 of them). The LLR score is simply a statistic; the higher the number the closer the candidate is to being a collocation. A trigram analysis could yield terms such as “electronic travel aid”, “object detection system”, or “low vision rehabilitation” which could be interesting for analysis. However we feel that a second-order analysis of terms may not necessarily lead to more insights or novel terms.

Community detection is a popular way for uncovering important structures and functions of complex networks. We performed a modularity-based analysis of the word co-occurrence graph (publication data) in order to detect underlying community structures. The network analysis tool Gephi ¹⁷ was used to visualize and understand the graph. Our word co-occurrence graph regards words as vertices and the LLR score of word pairs as edge weights. We constructed a temporal graph in the GEXF file format retaining these information. The Force Atlas algorithm was used to produce a layout with the strongly connected nodes pulled together (Fig. 6). Gephi’s Modularity statistic was used to detect communities. Force Atlas algorithm and the Modularity statistic are described more fully in the Gephi software’s documentation [9].