"Beeline" has developed a specialized neural network that will help in the search for the missing: the platform is called "Beeline AI - People Search".
The solution is designed to simplify the work of the search and rescue squad "". Since 2018, this team has been using unmanned aerial vehicles for search operations carried out in the forest and industrial areas of cities. However, the analysis of images received from drone cameras requires the involvement of a large number of volunteers. Besides, it takes a lot of time.
The Beeline AI - People Search neural network is designed to automate the process of processing photos. It is claimed that specialized algorithms can reduce the time for viewing and sorting the received images by two and a half times.
The platform uses convolutional neural network technologies, which increases the efficiency of machine vision tools. The neural network was trained on real sets of images. Tests have shown that the accuracy of the model on the test images is close to 98%.
The primary task of "Beeline AI - People Search" is to sort out "empty" and non-informative photos that definitely do not have a person or attributes indicating that there was a person in this place. This allows the analysis team to immediately focus on potentially high impact images.
The system can adapt to different conditions. It equally accurately locates objects both from a height of 30β40 meters and from a flight altitude of 100 meters. At the same time, the neural network is able to process images with a high level of visual "noise" - trees, natural landscape, twilight, etc.
βPotentially, the neural network is able to find people and objects in all search areas, such as forests, swamps, fields, cities, regardless of the time of year and the personβs clothes, since the algorithm is configured to work at any time of the year and can potentially recognize a non-standard body position in space, for example, a person sitting, lying or partially covered with foliage, βBeeline notes.
Source: 3dnews.ru