At the time of the outbreak of hostilities in Ukraine, 3 sets of sensors were manufactured.
One set of sensors was in Oleg Leonidovich’s backpack, the second set of sensors was in the office, and the third set of sensors was at the stables.
And when on the 24th of February the shelling of Kharkiv city started, the office was assembled and the rapid evacuation of people from the combat zone began.
It was necessary to make decisions and move very quickly. In such an urgent situation thoughts about saving the sensors were of secondary importance. Therefore, a set of sensors that was in Oleg Leonidovich’s rucksack miraculously got under the evacuation. Thus, only one working last prototype of sensors was left.
Part of the team headed by the head was evacuated to Chernivtsi. At that time it was already realised that the parent company was destroyed and would not be able to function in the near future. Kharkiv was in the zone of active hostilities.
We had to interrupt testing for some time, the tasks of evacuating women and children came to the forefront and the project team was fully occupied with these tasks.
However, we could not abandon the project.
Therefore, a new base for testing was found. It was a private small stable for 8 horses. But it was in a terrible condition. The team took part in the improvement of the stable. Within a fortnight the team managed to transform the stable, for this they were allowed to test the equipment on the existing horses.
Between 3 and 4 training sessions were recorded per day. Video of the training sessions (available) The horses that took part in the testing and training of the sensors were from a series of simple pets.
Thus, work continued to collect data and form the basis for the neural network.
During this period, a task was started with Gilberto to identify organic or non-organic movements of the horse. Why? Because the variance of movements is very wide. Each movement of a horse is different from the previous one. But there are movements that can be repeated to a certain extent. And there are movements that are completely out of character. Plus these movements are very often shown when the rider demands from the horse movements that are not quite characteristic for him.
The kind of movement that a horse in a normal situation does not do in nature.
This was the process of data collection, then the data was imported from the microSD, then the data was converted into a CSV file. After that the data was sent to Gilberto analytics.
But the next question arises – for each horse there are 4 sets of data – how to combine them? And the next task appeared – recording on microSD must be synchronous. The sensors should start synchronously. Dmitry Oparin wrote the following firmware, which allowed to combine these 4 sensors on the legs. It became possible to start recording on the start of any of the sensors. When you press one button, all 4 sensors start recording the dataset at the same time.
This was the next step in the development. From this moment we started to get synchronised data for all 4 legs. We now had a conditional description of a solid horse.