Kode Dot Brings Computer Vision to the Palm of Your Hand

To run a computer vision algorithm, one would typically use a powerful workstation or some type of specialized hardware solution designed specifically for this type of work. But these days, even the smallest of computing chips can get in on the action. Redditor luismi_kode’s recent project, for instance, demonstrates how an ESP32-S3 microcontroller can run a well-known edge detection algorithm in near real-time.

For this project, luismi_kode used a Kode Dot as the base. This device provided the ESP32-S3 chip and a 2.13-inch AMOLED display all in one little package. An OV2640-based camera module was added to the build to capture a stream of images. A Sobel edge detection algorithm was coded up to find all of the edges present in the incoming images. Those edges are displayed on the Kode Dot’s screen.

Sobel edge detection is not exactly in the same class of algorithms that, for example, power a self-driving vehicle. Moreover, the images were downscaled to a resolution of 160x120 pixels, and were converted to grayscale prior to processing. That significantly reduced the computational complexity of the problem, and by extension, the computing resources needed to run the algorithm. Even so, watching the ESP32 chip crank out edge detection results in near real-time is quite impressive. Kind of makes you wonder what else an ESP32-S3 can do, doesn’t it?