This project aims to develop a camera based system to provide assistance to the blind. Using glasses with a camera allows the hands to remain free while speakers in the classes provide voice feedback on objects or text identified by the camera. The system utilises the processing power of an android smartphone, reducing the power consumption and cost of the glasses.
Tensorflow lite provides an object detection example for android here which we used to help us integrate object detection into our android app. In this video we show video streaming from the ESP32 over Bluetooth, and object detection correctly identifying a cup.
As the next steps we want to add text detection, and optical character recognition, to allow things such as food boxes to be identified.
The aim is to use an ESP32 for image capture and data transmission to the smartphone using Bluetooth.
In order to ensure that the data rate was acceptable, a proof of concept was created using an ESP32 camera dev board. We could not find any examples of Bluetooth camera data transfer using the ESP32 but we were able to produce a simple demo based on the Bluetooth serial port example provided by Espressif. Using the esp32-camera library, we were able to get a quite reasonable frame rate over Bluetooth to an android smartphone at VGA resolution. I have made the Bluetooth camera code for the ESP32 available here.
Code for receiving the video stream on Android is available here.