Automatic insertion of advertisements into a video sequence

This paper proposes a method for seamlessly inserting virtual advertisements into a video sequence. The algorithm is based on projective geometric invariant for feature tracking that predicts the feature´s location at the next frame under perspective projection. The cross-ratio of points that extracted in pervious frames will provide the geometric constraint to predict the feature position in the frame even the feature is lost in some frames. The candidate regions for an advertising image in the video sequence are automatically determined by its structure and robustness. The experimental results demonstrate that the proposed method is effective and easy to insert advertisements into the video sequence.