Probabilistic appearance based object modeing and its application to car recognition

This paper describes a method for object detection and recognition based on appearance based approach. We introduce a probabilistic model to describe the wide variation of object appearance in images. In our method, objects are modeled as probabilistic features of silhouette and edge. These features are extracted from the object images viewed from various distance and orientation, and form the training data set for template modeling. A Non linear template model is build by the combination of Principal Component Analysis (PCA) and Kernel Ridge Regression (KRR). Finally, the problem of object detection is formulated as maximum a posteriori (MAP) estimation using above model. Experiments are conducted on road surveillance, where our method is applied to a certain car type recognition.