A new object detection and tracking method is proposed based on generative and discriminative models. Firstl

stable features of object are extracted and represented by the DAISY feature descriptor

which has computational efficiency and is invariant to illumination

deformation

viewpoint

and scale. In this way

the object generative model is constructed. Second

the Hough forest classifier is adopted as discriminative model and the input patches of object are trained. Moreover

the discriminative codebook is updated by computing the similarity measurement between the detection results of the following video sequence and the codebook

and makes the codebook dynamic and adaptive. Experiment results show that the proposed algorithm

combining DAISY feature descriptor and Hough forest

has satisfactory tracking precision and good real-time performance

Additionally

it works well under the condition of partial occlusions and different image resolutions.