A collision detection algorithm for NURBS(non uniform rational B-spline)surface is proposed. The technique of inserting node in reverse is used in NURBS surface subdivision

and a method for the accurate collision detection using distance testing between control points is presented. In the algorithm

firstly

the top control points that have clear geometric meaning are inserted

then the inserted nodes are calculated in reverse

and the insert technique is used to divide the curved surfaces finely; secondly

through building FDH(fixed direction hull)for the control points of curved surfaces that are finely divided by degrees

the part that is impossible to join in collision can be excluded early; finally

the distance testing is carried out if the subdivision level reaches a certain threshold. Through the collision detection between a butterfly and corn leaves in the virtual environment

it is verified that this algorithm can be used for collision detection between objects defined by NURBS surfaces. When the subdivision level is 5 and the threshold is 0.000 6

the collision detection in this experiment is real-time with good accuracy.