EPISODE 5. OPTICS AND THE SPECIFICS OF SHOOTING WATER

When it comes to computer vision in a pool or open water, the first thing to think about is not a neural network, but optics. The camera is the eyes of the system, and whether the algorithm can recognize anything useful at all depends on the quality of the “input image.”

Let's start with cameras. In most cases, IP cameras are used for such tasks, which connect to the network and transmit a video stream in real time. Here, a wired connection has a clear advantage: signal stability. Wi-Fi is convenient, but it is prone to delays, packet loss, and congestion. When every second can cost a life, connection reliability is critical.

But it's not just data transmission that matters. Water creates a complex picture: sun glare, reflections, flickering shadows. To combat this, polarizing filters are used to reduce reflections and make the surface more “transparent.” This is a simple but very effective technology familiar to photographers and cameramen.

In addition, cameras are now used that can shoot not only in the usual visible range, but also in infrared or even thermal. The infrared range helps to “see” a person when the lighting is poor or the water surface causes too much glare. Thermal imagers can detect the temperature contrast between the body and the water.