OBJECT DETECTION Using ESP32 Cam

DO-IT-YOURSELF: PROJECT

- SOMNATH BERA

An ESP32 camera (cam) can take pictures at 2MB per second rate and transmit them up to 200 metres using a small antenna on 2.4GHz. But when it teams up with an internet-enabled network it can do artificial intelligence (AI) related work as well.

In this project the ESP32 cam takes and sends the pictures to a remote computer that is connected to the internet, pulls the specified AI model, and analyses the object (model) to identify it. It then sends the object’s identity back to the ESP32 cam, which then executes the object-specific command entrusted to it.

When the detection computer is on, the ESP32 cam is all vigilant and ready for action. Once the model is loaded (identified), an active internet connection is no longer required. However, the computer needs to remain on all the time.

The ESP32 cam also reproduces the live feed from the ESP32 camera. The IP address is set as fixed type (192.168.43.11), which can be changed as per requirement.

The ESP32 continuously draws the image boundary over the image. If it finds two images, it draws two boundaries, and so on. Its confidence level can be set from 0 to 100%.

Image Resolution, MirrorImage, Flash (on 100% switch on the in-built LED), Quality, Brightness, and Contrast (see Fig. 6) can be changed. However, on increasing image resolution from 320×240 to 1600×1200 the system become extremely slow. It is better to leave the control at the default level and change only if necessary. Author’s prototype is shown in Fig. 1.