A comparison of LIDAR waeveform decomposition models

The storage capability of full waveform is the state-of-the-art technology of LIDAR. In addition to the 3D point information recorded by conventional LIDAR systems, waveform LIDAR systems encode the intensity of returned signal along the time domain. This provides a user the possibility to decompose the waveform for the detection of illuminated target points. Therefore, the locations of the illuminated targets can be refined by analyzing the waveform data. So far, a standard approach to waveform decomposition is still not available. A waveform may be a composition of some prominent, overlapped and weak return pulses. A waveform decomposition method may easily detect a prominent return pulse, but it usually has some difficulties to deal with overlapped and weak pulses. In this paper, a waveform fitting method which takes the overlapped pulses into consideration will be applied. To fit a waveform, the number of returns is needed. The initial number of returns commonly is determined by a simple way such as the number of local maximum of a waveform which is influenced by noises and overlapped echoes. For this reason, the wavelet transform is used to estimate the initial number of returns in this research. Moreover, compared with taking Gaussian as the fitting basic model, another basic model, log-normal is included. Our preliminary results show the effectiveness of wavelet transform to determine the initial number and the ability to detect overlapped pulses. The Log-normal model has better fitting results than Gaussian model in forest areas. As the results, the points extracted by our developed methods increase compared with the points extracted by commercial system. The increased points can be useful to future applications especially in a forest area.