This repo is about the project i did in 2019 summer robot and AI of imperial college London. I wrote it for realsense camera D415 in ubuntu18.04. By the time 2019-8-10, Realsense camera python SDK only stably support Linux and Windows. So i think it may not work in Macos .

This code is wrote in python3.6

It mainly consist of three part.

1. calibration part: using the opencv findChessboardCorners function and depth information to fast calbrate camera.

2. QR-code tracking part: i use qrcode to generate QR-code, and pyzbar to recognize every frame, then use kalman filter to stabilize the tracking

2. NN model part: i use Segmentation-driven 6D Object Pose Estimation(CVPR 2019) to predict the pose of the objects in RGB image

1. create a python3.6 virtual environment with pip and enter it
2. enter the LachFore fold, type pip install -r requirements.txt
3. install realsense SDK ,connect to the realsense camera, you can use realsense-viewer to see if it is connected
4. git clone https://github.com/cvlab-epfl/segmentation-driven-pose.git, download pretrained weight following the instruction of the github repo.
5. python main.py

1. most of the parameters are in world.py
2. you can change the model easily, you should rewrite the class in predict.py , which should have predict method
3. i draw lessons from realsense code examples a lot, like the realsense_device_manager.py , but i change it for only one camera situation.