Project presentation “Deep neural networks for street view type images” from the Sectoral Fund CONACYT-INEGI.
The Sectoral Fund CONACYT-INEGI grants support and financing for the realization of scientific investigations, technology development, innovation, and national or international registry of intellectual property.
The principal objective of the seminar was to present projects, advances, and principal results of funded works for the Sectoral Fund to exchange ideas between projects developers and potential users.
Hunabsys R&D, represented for the CEO José Carlos Castro and Fernando Carrizosa, present the “Deep neural networks for the identification and extraction of geographical objects in street view type images” project. The principal objective of the project is to automatically generate information from various elements of urban infrastructure and equipment.
The prototype is based on a GPS-RTK sensor that uses signals from the Global Satellite Navigation System and a reference station to correct ambiguities in the position calculation improving its accuracy. In addition to having a GPS-RTK sensor, the prototype includes a magnetometer to calculate its orientation. The prototype consists of an outdoor NEMA box where all the electronics are located, in addition, it has a 6 camera circular array that captures 360º images.

Regarding the software, several packages have been implemented in order to obtain the geo-referenced information of the territory, e.g., Labelme, Caffe, Arduino, Python, OpenCV, among others.
The project is based on the use of artificial intelligence via deep learning algorithms to identify the desired objects.
In order for a deep learning algorithm to “learn” how to perform a task, it needs to be trained on a computer by using images where the elements of interest are indicated, in this case, these are objects of urban infrastructure such as traffic lights, luminaires, signs, etc.

Complete video of the “Deep neural networks for the identification and extraction of geographical objects in street view type images” project presentation.