I’m currently working on a pan-tilt camera project where I intend to be able to detect objects on the screen and then when a person comes up it should track and follow the persons movements. I am currently working on the object detection side of the project but I am trying to see how I can check and see if a person is present in the frame then it should follow the person till the person goes out of view. I am using servo motors but i’m not entirely sure of how to implement it after training and deploying the trained model to raspberry pi 4.
If you deploy a EI FOMO Impulse, then the inference will return the rough position of the Bounding Box within the image. Using the position data, you can draw an icon such as square, rectangle or drop a pin on the person in the image. On the next inference, erase the previous location icon on draw a new icon. Based on the icon’s position you can then pan/tilt the camera to place the predicted new location of the person in the center of the camera. IF your inference times are low say under 300 ms, then observer will not notice a slightly mis-placed icon since it will be over-written quickly. You’ll use the current position and previous position to draw a prediction vector to move the camera.