TLS data denoising by range image processing

Abstract

The work described in this paper deals with denoising of fine detail in terrestrial laser scanner (TLS) data for close‐range applications. In contrast to other denoising methods described in the literature, the method described here puts the problem back into two dimensions. In the proposed method, data from each instrument station is processed by denoising the range image as a 2D function. Then, after denoising, the registration process is applied to obtain the final 3D point cloud. Two image denoising methods are tested. Wavelet analysis and a “non‐local means” (NL‐means) algorithm are applied to archaeological objects of sizes described as medium (around one cubic metre) and small (around one cubic decimetre). The rich relief of these objects proves the efficiency of the method on all kinds of surface shape. The experience shows that the NL‐means method achieves good results: the standard deviation of 2 mm in the raw data is decreased to 1 mm. Besides, NL‐means denoising requires very few parameters to be set (generally two) whilst wavelet denoising requires many. Hence, the extreme simplicity of the NL‐means method makes it much more efficient and suitable for use by non‐specialists. The method, which slightly modifies the standard production process and inherits the algorithmic complexity of the classical 2D image processing schemes, allows the range of applications of TLS to be extended to small and finely detailed objects.

Evaluation of Wavelet and Non-Local Mean Denoising of Terrestrial Laser Scanning Data for Small-Scale Joint Roughness Estimation

“…As proved in some recent studies (Bitenc et al, 2015a;Khoshelham et al, 2011;Smigiel et al, 2013Smigiel et al, , 2011Smigiel et al, , 2008 range error can be successfully reduced by image denoising methods. The TLS denoised surfaces show details, which may otherwise be lost in noise.…”

“…Once the complexity of a 3D randomly scattered point cloud is reduced by gridding to a regular 2.5D surface, a wide range of existing image processing algorithms can be used. An overview of image denoising methods and further references can be found in (Buades et al, 2005;Smigiel et al, 2011;Zhang et al 2014). In this research the two methods (DWT and NLM), together with their two variants, are tested for a reliable roughness estimation.…”

“…In our previous study (Bitenc et al, 2015a) denoising results of Discrete Wavelet Transform (DWT) were analysed, where different DWT methods as well as thresholding procedures were tested. Smigiel et al (2011) compared the DWT and Non-Local Mean (NLM) denoising methods for a surface reconstruction of a medium (around one cubic meter) and small (around one cubic decimetre) archaeological objects. They advocated the use of NLM.…”

Terrestrial Laser Scanner Data Denoising by Dictionary Learning of Sparse Coding

“…We have first investigated the wavelet approach. We have then showed that the NL-means approach is much simpler to implement and more efficient as well (Smigiel et al, 2011). In this paragraph, we show very initial results of denoising based on dcitionary learning of sparse coding.…”

Evaluation of Wavelet Denoising Methods for Small-Scale Joint Roughness Estimation Using Terrestrial Laser Scanning

“…An advantage of 2.5D surface is that a whole range of existing image processing algorithms can be used. An overview of image denoising methods and further references can be found in (Buades et al, 2005;Smigiel et al, 2011;Zhang et al 2014). …”