You might find this post relevant to your goal:
Steering Behaviors - Obstacle Avoidance - #7 by jeremydouglass
Your current post have three different questions. I recommend you keep on question per post so each question gets a proper deserved answer.
For the first question I suggest the next exercise. Draw a big blue circle in the middle of the sketch simulating a person (as if it was detected by a kinect unit). Then with your mouse, you click+hold anywhere in the scene. This defines the initial position of the fish. Then you drag and release and that will define the velocity and direction the fish will travel with respect to the starting point. Your third task would be to implement a code for collision avoidance. What do we need for this operation?
Some thoughts: You will need to determine the distance between the “fish” and the person. Since the person is a single blob, you should think (or do) the circle to be filled with many points. Next, how would you determine the distance between the fish and the person? Would you use the center of the person? Or would you use a point in the contour of the person? (Please ask if you don’t know or you are not sure)
An additional note. If you mix the depth and the image, you will get a 3D space. Would the fish move in 2D or 3D? I suggest to stick to 2D and explore the power of Kinect for “person/people” detection. After you understand this concept, you can move to the 3D approach or you could try to implement the sphere (second question). For the third question, you will need to experiment with the depth/image data. Resizing an image is a common task in Processing. However, I can see some challenges resizing the depth data provided by Kinect after is mapped to the image. Maybe it is a simple transformation. More research is needed plus probably access to some sample data.
Finally, you might find this useful while exploring this technology: https://shiffman.net/p5/kinect/
Kf