4. Vision Subsystem : Performance Evaluation

 
 

4.1 Experimental Design 

  The memory and processing footprint of the image processing software are basically negligible.  Testing merely involved running the program and then using the Linux command ps aux to determine the memory usage.  As far as timing the program, the results here were basically just determined by "eyeing it."

4.2 System Statistics

  The system used to perform these test is a dual Pentium III 500 with 256 MB of RAM running RedHat Linux.   

4.3 Test Results

    4.2.1 Run Time
  Run time consists of two parts: the calibration, which must only be done once at the beginning of each game, and processing the before picture, after picture, and there difference for every move.  Processing the calibration image takes between a second and a second and a half (it is almost exactly the same every time).  Likewise, the before picture takes about a second or less to process, as does the after picture.  Finally, the difference between the two and the final processing takes about a second.  Altogether, for each move the total processing time for the vision aspect (excluding the time it takes for the operator to take the picture) is probably less than three seconds.  This doesn't vary noticeably, because the pictures are the same size every time, and the same algorithms are used every time.   

 

     4.2.2 Memory Requirements
  The memory used by the savant program at its peak memory usage is only about 3.5 MB of main memory.  This is when both pictures are loaded and the difference is being determined.

 

[Previous Section] [Table of Contents] [Next Section]