ESP32CAM reading direction indicator

Help
#1

Question/Issue: I want to use/train an ESP32CAM for reading a simple direction indicator (green square) but I’m too overwhelmed to know where to start

Project ID: 2 wheeled robot for educational purposes

**Context/Use case: I’m a teacher and I have recently followed a workshop with my students in which we trained a model to recognise cars. This model was loaded in a ESP32CAM which opened a gate when a matchbox car was be put in front. After this workshop, we came to the idea of using an ESP32CAM in our 2 wheeled robot. This robot is able to follow lines but there are T junctions in the lines. There is a colored square at each T junction which makes it possible for the robot to know which side it should take. We 'd like to replace the color sensor with the ESP32CAM. This makes that there are 3 possibilities: no indicator found, left turn indicator found and right turn indicator found. It is not so difficult to get training data (e.g. video footage from the robot) but apart from that, I don’t know how to start. Any advice is welcome :slight_smile: **

#2

Hi @GuyBuys,

It sounds like you need to use some form of object detection, as you may have one or more indicators in your field of view. I recommend taking a look at this talk (tinyML Talks: Constrained Object Detection on Microcontrollers with FOMO - YouTube) to learn more about how FOMO works on microcontrollers. Here is an example project demonstrating FOMO on an ESP32 to identify beetles (FOMO Edge impulse detecte potato beetle - Hackster.io).

From that, you can train sample images to look for your indicators. Once the indicators are identified, you can use their location (i.e. left vs. right side of the image, top vs. bottom side of the image) to determine if your car needs to drive forward more and when to take the turn.

Hope that helps!