Abstract:Existing scene recognition system often needs a lot of scenario training data for training, to collect the data is often difficult, and the training is offline. For new training, new scene is needed to add, so the system real-time performance, scalability, and robustness are poor. A scenario online learning method based on incremental principal component analysis (PCA) was presented. Through incremental PCA algorithm with subspace realtime update capabilities, the sample projection of PCA was calculated, and two discriminant θclass、θdistance were set up to handle different sample situation to reduce the amount of calculation, to realize aims of incremental online learning and recognition a sample scene. Experiments show that this method effectively solves the difficulties of training data collection, implements the scene knowledge accumulation and online update, and greatly enhances the PCA algorithm of real-time performance, scalability and robustness.