Investigating Full-waveform Lidar Data for Detection and Recognition of Vertical Objects

A recent innovation in commercially-available topographic lidar systems is the ability to record return waveforms at high sampling frequencies. These “full-waveform” systems provide up to two orders of magnitude more data than “discrete-return” systems. However, due to the relatively limited capabilities of current processing and analysis software, more data does not always translate into more or better information for object extraction applications. In this paper, we describe a new approach for exploiting full waveform data to improve detection and recognition of vertical objects, such as trees, poles, buildings, towers, and antennas. Each waveform is first deconvolved using an expectation-maximization (EM) algorithm to obtain a train of spikes in time, where each spike corresponds to an individual laser reflection. The output is then georeferenced to create extremely dense, detailed X,Y,Z,I point clouds, where I denotes intensity. A tunable parameter is used to control the number of spikes in the deconvolved waveform, and, hence, the point density of the output point cloud. Preliminary results indicate that the average number of points on vertical objects using this method is several times higher than using discrete-return lidar data. The next steps in this ongoing research will involve voxelizing the lidar point cloud to obtain a high-resolution volume of intensity values and computing a 3D wavelet representation. The final step will entail performing vertical object detection/recognition in the wavelet domain using a multiresolution template matching approach.