Abstract: Traditional vision tracking algorithms cannot track blank paper very well since blank paper lacks significant texture details leading to insufficient feature points. And the situation becomes worse when the tracking object moves with large rotation. To solve these problems, an improved compressive tracking algorithm with structure constraint is proposed. Firstly it uses line detection to find the line segments of the tracking samples, and then measure the angles among the found line segments to remove those obvious wrong samples with similar feature values. On one hand, the structure constraint reduces the sample search space and enables the compressive tracking algorithm to track objects with large rotation; On the other hand, the four exact corners of the blank paper are tracked at the same time, which are used to locate the position of the whole paper. The experiments have fully confirmed the validity of this algorithm, illustrating it can track well for such blank paper without texture when the object undergoes large rotation.