Santa or Not Santa Using a Raspberry Pi and Deep Learning with Keras and TensorFlow

Computer vision is a one of the most fascinating topics in robotics, and tech in general, today. The entire field can be reduced down to one very simple question: how can you teach a computer to recognize objects in the world around it? The answer is that you can’t—at least not in the way we intuitively think of it.

When a computer looks at a video feed, all it’s really seeing is a bunch of pixels organized into a series of frames. And yet, your smart phone can recognize your face. Security systems can recognize the outline of a gun in a scan. Satellite systems can find ancient ruins, by seeing anomalies in the surrounding geometry. How does this work, if the computer itself doesn’t have the intelligence to understand the elements it’s seeing?

They do it be recognizing patterns. A face consists of two dark spots representing the eyes, with a line representing the mouth, and at fairly consistent proportions. In the early days of computer vision (CV), these patterns had to be taught, but now a computer can learn them using deep learning and neural networks.

One of the foremost authorities on CV (particularly on Raspberry Pis) is PyImageSearch, and they’ve got a fantastic tutorial explaining these concepts in a holiday friendly way. This tutorial explains how a neural network system (Keras and TensorFlow) can learn to recognize anything—in this case Santa Claus. When the system detects a Claus, it lights up a Christmas tree. They provide a pre-trained neural network for the project, so you can build this yourself in no time.