Roboticsinfo,
Ok, let me explain these things again. It is better to use 8 IR sensor to cover the entire robot. The reason is that the robot will move to any direction the sensor detects this, and the object must be inside the range. The reason for using 8 sensors is BECAUSE Infra Red is a light, photon, and it follows the optics. In other words, they detect a single point where the photon gets hit and it gets bounced back to the IR receiver. So, you have multiple points but to cover the corners u must use another points for that. <<That's why I suggest 8>>
The range of the most IR sensors cannot exceed 59" . This is the maximum. Where did u get the information? over 59"? Unless you invented your own IR sensor this will be possible. The most popular IR sensor is Sharp GDs series
https://www.acroname.com/robotics/parts/sharp_guide.pdf
The sonar is better because the wave gets scattered from the source so it provides 23 degrees of sonar vision. Yes, you can take sonar pictures using this information.
The CMUCam is the easiest if and only if the person wears one single color that is unique to the environment. Otherwise, you will need to use your own code and reprogram the CMUCam microprocessor. Simply, to detect a person without color detection, you have to use two different techniques in computer vision; Template comparison, where you have a close shape to that person and you compare it to the picture, and Edge detection ,where you differentiate the edges of that person. << That’s why I said it is away so advanced, you need at least to use a lot of algorithms and you must have a solid knowledge in computer programming>> Also, I want to add one more thing about color tracking of the Cmucam alone is useless because not all the objects around looks exactly in the same color. There is always color intensity around us. To solve this, you must use a halogen flash light , like the one in your camera. All the colors will look exactly the same.