Abstract: A roof detection method using LiDAR data is proposed based on points' normal with weight.The normal of the plane,which is composed of point and its neighbor points,is counted to find out the high frequency,and the roofs are detected.Each point's weight is considered to determine the contribution to the plane which diminishes the affection of noises to some extent,and increases the accuracy of the small roof detection.In addition,multiple thresholds are adopted to detect roof planes with different sizes.Experiments are given to prove the algorithm proposed.