• Capstone Home
Senior Project
• Overview
• Examples
• Downloads
• Journal
• Timeline
• Requirements
Professional Info
• Career Path
• Philosophy Statement
• Resume
• Contact Info
|
Project Journal
|
04-30-2006
This will be my final journal entry. My presentation last Tuesday went extremelly well! This project has been a huge undertaking, but has really been worth it. I learned a lot and enjoyed the opportunity to put my computer science knowledge to work on a challenging, unique, real-world problem. Although I probably didn't meet any of the deadlines on my timeline, the results speak for themselves! The Zoom was flawlessly controlled for over 1 mile on a rural road by my vision algorithms! I believe that if that road had been 500 or more miles long, of similar topology, the robot would have been able to go all the way from one end to the other without human intervention (although it might need a bigger gas tank!) - I couldn't ask for a better feeling of accomplishment from a senior project! This was truly more than "just a project!"
04-20-2006
This update has been a long time coming. A lot of thought process, number crunching, geometry / vector graphing and finally coding to make it all a reality has been accomplished since the last entry!
Major revisions have been made to the core lane detection algorithm. The abstract line generation phase has been improved to generate the proper harmonically spaced rays, as illustrated in the examples section, according to proper geometry rather than the somewhat "makeshift" method that was in place previously. Also, the line scoring routine has been improved so that every line is scored in a consistent manner (same frequency of pixel checks) regardless of its slope - this was not the case up until now. These two factors have dramatically improved the accuracy of the lane detection algorithm.
The lane analysis phase has also been improved to not only consider the current trajectory of the vehicle when determining a correction value, but also the horizontal position of the vehicle within the lane - this was the "missing link" and has made an enormous difference - the algorithm now consistently performs far better than it ever has in the past!
The latest road test of the Zoom vehicle, using these updated vision algorithms as the sole navigational mechanism, successfully guided the unit, at a high rate of speed (for a lawn tractor anyway), for a distance of at least 1 mile without faltering! Even more impressive results are expected in the near future!
03-04-2006
I started working on my paper this weekend. I have a title, abstract and outline. Now I'll just be taking it one paragraph at a time!
02-19-2006
Waiting for the weather to improve so Mr. Blahnik and I can do some more live testing!
02-14-2006
Today I finally got a chance to give Dr. Pankratz a demo of my software, both the lab and robot interfaces. We talked more about the possibility of using statistical analysis to effectively eliminate the already small percentage of inaccuracy encountered by my algorithms. In robot mode, combined with a higher frame rate, it seems possible to achieve > 99% accurate heading corrections.
02-12-2006
I once again spent many hours coding this weekend. I made a some really important progress: I integrated my algorithms into shared Visual Basic module files that are cross compatible with both my frame-by-frame testing interface "Lab Mode" and the Zoom's control program "Robot Mode". From this point forward, when I update my algorithms, which always happens based on lab mode testing, no additional labor will be required to be able to test them on the robot, since both interfaces reference the same files for my algorithms. This is a really big step and will allow for more frequent testing on the robot!
02-09-2006
I met with Dr. Pankratz and filled him in on my current progress, especially regarding methods of objectively measuring the accuracy of my algorithms. He offered a good idea: use an existing set of test images and manually insert the road boundaries into them, then gauge the accuracy of my algorithms based on how close their lines come to the pre-determined ones. I really like this idea and will probably try to implement it along with a batch processing feature that I have been thinking about.
02-07-2006
Mr. Blahnik and I conducted some tests on the Zoom unit to 1) Determine the optimal servo positions for neutral (the left and right side do not behave synchronously) and 2) Determine the relationship between the range of servo positions and the range of wheel speeds (This is not linear and is also asynchronous between sides). Through statistical analysis of this data, we are working to develop a table of relative wheel speed vs. servo position, for each side, so that the guidance software can exercise more precise and predictable control over the movement of the robotic vehicle.
02-05-2006
I spent a long time working on my vision algorithms this weekend. The frame-by-frame analysis program is now MUCH more detailed and provides analytical data that will be helpful in the quantification of accuracy as well as self-correction routines. I fixed the "score line" module of my algorithm so that it correctly assesses laderal pixel hits based on parallel line equations. I improved the "line rationalization" procedures to help the algorithm better decide whether or not to "trust" a line -- this is somewhat based upon the analytical statistics feature that was added. I also added several image pre-processing algorithms in order to provide better input for the edge detection phase. Also added some auto-calibration routines for various algorithm variables. Overall, made A LOT of progress! Hopefully the next tractor test will be even better than the last!
01-31-2006
Mr. Blahnik and I conducted a test run of the tractor in order to observe 1) The implementation of my smooth turning algorithm and 2) The efficacy of the current version of my lane detection and analysis algorithm as the sole navigational aid for the tractor. The results were better than expected and very encouraging!
01-30-2006
I had a long meeting with Dr. Pankratz today in order to make sure that we were both on the same page in regards to the scope and current status of my project. This went very well and DCP came up with some good ideas on methods to quantify "accuracy" in order to measure the effect of changes to the algorithms. I will probably try to implement them soon, developing a more controlled environment for testing, so that I can more objectively monitor the success of ongoing algorithm development.
|
|