Vision Using Web Cam

I've been playing with RoboRealm (RR) the last week to add vision to my Create-based robot. I have working the ability to seek out a red dot. The robot will get to a certain distance and then stop. If the dot moves away or closer the robot adjusts its position.

Next I started working with obstacle avoidance. Basically how can you find a clear path through a room? The RR site has a number of examples which I tried. They work fine on static images but with live images they start having problems. I also found problems with most of them in real-world situations. This isn't even having the robot actually move but just analyzing my office and putting some obstacles in front of it. (You want a real nightmare? Put a wicker waste basket into an image being processed for Canny edges. It looks like worms crawling all over the place.)

One issue is that pixel changes from image to image in the live feed can cause significant changes in the end result. I'm wondering how much this is caused by the web cam. I realize there are going to be pixels with borderline color values that can toggle between those values. But I see pixels twitching in areas where the color is consistent for the area.

Are some web cams more stable than others?