Machine Learning Will Help Improve Google Maps Services

Finding places could have been much difficult if one is deprived of super handy mobile app Google Maps. It has undoubtedly made our travel experience undeleterious. Be it direction, traffic updates, business listings, public transit information, Google Maps is probably the one app that we use every time while wandering around.

The Search Engine Giant Google often updates Maps service with new addresses and information. That said, company has now decided to take a step further by incorporating some deep-learning tech i.e. Machine Learning to optimize the Google Map Experiences.

In a recent blog post, Google has stated that it has started to implement the deep learning techniques to the Street View as well as the Google Maps to help map with new addresses. Google is now likely to use the photographic data that Street View cars collect while their trip to extract additional place information such as street names and numbers etc.

One of the major problems encountered by the app system was that the photos that the Street View cars captured were not clear enough for places to be depicted. More often, it had been observed that there were some discrepancies in the angle of the shot, obstacles, distortions, that made it difficult for anyone to comprehend the names and numbers.

Even, Google’s ReCaptcha solution also hit a roadblock as it is almost impossible for human being to look and manually analyse over 80 billion of  hi-resolution photos collected by the Street View Cars. As a result, they looked for something to solve the problem. As a solution, Google has finally infused the artificial intelligence and the deep learning to automatically fetch information for the geo-located images.

How does it work?

The whole idea of how this system work is that it makes use of the deep neural net that makes the image information reading process automated. The latest Machine Learning algorithms have helped achieve an 84.2 per cent accuracy when tested on several challenging street signs in France. These stats significantly outperformed the previous state-of-the-art systems.

Google has also made this algorithm publicly available on GitHub through TensorFlow which is Google’s own open-source machine learning software library.

Furthermore, Google has already been implementing machine learning to indistinct faces and car license plates. And now, it has begun to use the same technology to fetch information from the street signs. Google believes that they could improve the location data of about one-third of the world’s addresses using this technology.

In this move, the company not only improved the software to read only the street numbers but also the street names. The new algorithm can even replace the abbreviations with full names and get rid of any irrelevant text in its photos.

This algorithm will also help map newly constructed streets and buildings that are not yet the part in the official maps of the cities. Moreover, it can save the names of the businesses enabling Google Maps to update useful information on business listings.