I am new to computer vision, and need some advice on where to start.
I am looking to track a moving robot using a stationary, external camera positioned on the ceiling. The robot should complete an obstacle course autonomously using feedback from the camera. In essence, I am looking to develop a "map" of where the robot currently is, which the robot would interpret and decide which direction it should go.
I have looked through many scholarly articles on the subject, but most research seems to be focused on cameras onboard the robot. I am assuming that using a stationary camera would be easier to track motion, and was looking for any advice on where to begin and the general steps I need to take (ex. camera-calibration, image processing, relay signal to robot, etc).
Thanks a ton!
As @Callari mentioned your question is not so clear. But I would assume two things for writing my answer: (1) Your environment is static, (2) Obstacles are also static.
You need to create a map of your environment from the camera. Use this kind of an approach to calculate the positions of different obstacles with respect to some reference point.
Once you have the positions of obstacles you need to do the mapping of your environment. If you have a metric map then it's nothing but an occupancy grid. This map has to be with your robot and if it starts from the same reference location, initialize your robot with some start_location and start_orientation. Use Dead Reckoning for localization and keep updating position and orientation of your robot as it move through.
Then you need to do the path planning for your robot. You can use A-Star algorithm to find a path for your robot. After reaching the goal point do the path planning again.
In case of dynamic obstacles, iteratively find positions of obstacles from camera and do the path planning. I am new to the subject but have gathered some information lately, due to my interest in robotics.