Filtering is an important method to extract digital terrain model（DTM）form Light Detection and Ranging（LIDAR）data and has made some good effects in application

but existing filtering methods are deficiency in certain applications

especially unsuitable for undulated terrain.In this paper

a new filtering method which is based on contours is developed.The method mainly consists of three steps:Firstly

digital surface model（DSM）is generated from raw LIDAR data by surface interpolation methods

such as the nearest neighbor interpolation method

and then contours of DSM are generated.Secondly

these DSM contours are separated into three groups:ground contour

which only consists of points on ground;object contour

which only consists of points off-ground

such as on trees or roofs;and mixing contour

which consists of points on ground or off-ground.According to some properties of contours

including closeness

start-to-end distance

length of curve

denseness

ground contours and ground contour section of mixing contours are extracted and then some ground points are obtained

finally an initial approximation of the DTM is determined by interpolating those ground points.Then a refined DTM is obtained by an iterative method:comparing the approximate DTM with the original DSM and if the difference is below a predefined threshold the original DSM points are selected and made into DTM data points.Otherwise

the location is marked as no data;and then the no data points are interpolated using the selected DSM data.Comparisons between initial and refined DTM obtained by methods proposed previously is done.To calculate the accurate probability

the objects obtained by new filter is compared with standard objects obtained from aerial images manually.All the experimental results show that the proposed method is a simple and practical one needing little calculation

and can be used in undulating terrain area.