How to collect data via camera and label them in order to train tensorflow CNN? about tensorflow-on-raspberry-pi

[samjabrahams]

Hi Xu,

Glad that the previous answer helped you out! Unfortunately, I don't have a huge amount of personal experience with streaming in data from the Raspberry Pi camera, and can't give a concrete answer to this. Additionally, these sorts of personal project questions are more appropriate for Stack Overflow than as a GitHub issue, which should really only bring up questions or problems about the TensorFlow binaries themselves.

To try to get you started, what I'd try to do is try to find a way to attach the camera as well as the RPi to an RC car, such as the Pi-car. The goal would be for you to manually control the car (through some sort of obstacle course, like with traffic cones) while capturing images/video. While capturing video, you would also be capturing your manual inputs on the R/C car, and you would connect the two together to line up the R/C video with your controls. Then for training, the input values would be your video images and the output target would be the controls you entered at that time.

This Stack Overflow question/answer may get you started for the video side.

Good luck!