What are the ways to achieve good accuracy in support vector machine ? (MATLAB)

Recently I works on depth values of an image. Basically those images are stairs [up & down].I draw a vertical line upon those images and take the 1D feature based on vertical line coordinates.But When i use them in SVM I did not get good accuracy. When i tuning parameter[c,gamma] it sometimes yields good.But problem is that after changing train / test set again needs tuning.Sometimes gets better or sometimes not.Main point is i did not get a consistent result.In this case what procedures i need to follow to get good accuracy.Any suggestion/advise will be appreciated greatly.

[John D'Errico]

What are the ways?

1. Get good data. Good data begets high accuracy.

2. Learn how to use the tool.