The development of the solid-state LiDAR allows depth sampling in non-trivial patterns. Nevertheless, the technology is still limited by the quantity of samples per unit of time. There are multiple ways to utilize this sampling budget. Our previous work shows that adjusting the sampling pattern according to the scene can be highly beneficial for accurate dense depth reconstruction from sparse depth samples. A method for adaptive depth sampling, which imitates human decision making, is currently under research. Amongst others, Computer Vision and Deep Learning tools are used in the development of a robust algorithm