Abstract: An improved automatic incident detection algorithm，based on principal component classifier，is constructed.Principal component classifier is an improved two models classification technology.This method computes a direction from a dataset which can be seen as the normal direction of a hyperlane such that samples in one class can be separated well from the other by this hyperlane.The projections onto that direction can be used for estimating class-conditional possibility density function（pdf） according to Bayes rule.The algorithm can also be carried out in the feature space to deal with more complicated classification problem by mapping input data into a high dimensional feature space using kernel function.Finally the simulation results on I-880 freeway dataset are：DR is 100.00%，FAR is 1.82%，and the MTTD is 1.02 minutes.