Aimed at the problem of position and attitude estimation for UAV, a relative position and attitude estimation algorithm based on target features in image was proposed. The initial location of the target was obtained with Camshift algorithm. The feature points in tracking area based on the nonlinear scale space were picked up with KAZE features, which were used to match with the feature points of the source target. The exact location of the target could be obtained and the target could be tracked quickly in the picture plane. The estimation of relative position and attitude between the unmanned aerial vehicle and target was conducted in the body frame of axes. Some experiments were fulfilled for the verification of the algorithm. Results show that the proposed algorism has strong tracking and real-time performance. 