1. Radiometric correction and normalization of LiDAR intensity data

I propose and develop two approaches to improve the radiometric quality of airborne LiDAR intensity data during my Ph.D and postdoctoral studies. The first approach, named as Gaussian mixture modelling (GMM) based sub-histogram matching technique, aims to normalize the overlapping LiDAR intensity data. Later on, I propose an alternative approach that relies on the use of polynomial function fitted on the joint histogram plot, which is generated with a set of pairwise closest data points. Both methods can significantly reduce the striping noise found in the overlapping region of airborne LiDAR intensity data.

2. LiDAR Scan Line Correction (LSLC): noise removal for LiDAR intensity banding

I propose and develop a LiDAR Scan Line Correction (LSLC) to remove the systematic striping noise found in individual LiDAR data strip, which is commonly known as intensity banding. The cause of intensity banding can be ascribed by the improper alignment found between the laser receiver and transmitter, resulting in the backscattered laser beams partially falling outside the receiver’s field of view. Such a problem usually happens in linear mode airborne LiDAR systems operated with oscillating mirror.

3. Multispectral airborne LiDAR intensity correction

The latest multispectral airborne LiDAR system may suffer from a combined issue caused by the abovementioned effects, i.e. 1) intensity banding found in each of the individual laser channels, and 2) system and environmental induced distortions found at the swath edges (See above sub-figure (a)). Therefore, I implement the LiDAR Scan Line Correction in each of the individual laser channels to remove the stripe artifacts / banding effect (See above sub-figure (b)). Then, I also develop an overlap-driven intensity correction (OIC) to deal with the stripe artifacts found at the swath edges (See above sub-figure (c)). The OIC is capable of estimating the correction parameters in each of the overlapping data regions based on iteratively-reweighted least squares supported with M-estimator. [PDF]

4. Automatic extraction of highway light poles and towers from mobile LiDAR data

The work was funded by NSERC Engage collaborated with a local surveying and engineering firm, Tulloch Engineering Ltd. I propose and develop a LiDAR data processing workflow to extract highway light poles and towers from mobile LiDAR data. The workflow first includes 1) ground filtering to normalize the above ground features, 2) a DBSCAN clustering to group the data point clouds in 2D, 3) a decision rule classification to identify potential light poles and towers and 4) a final data cleaning mechanism to remove the ground points from the extracted light poles’ structure. The final accuracy yielded more than 91% with a case study along a section of highway 401, Toronto, Canada and the work can be found here.

5. SLIER: Scan line intensity-elevation ratio for water surface mapping

I propose and develop a new airborne LiDAR ratio index, named SLIER, to automatically extract the water surface. SLIER can be computed by dividing the standard deviation of intensity over the standard deviation of elevation along each scan line, collected by any linear-mode airborne LiDAR sensor. Over the water surface, airborne LiDAR data are always found to have a high fluctuation of the intensity and a low variation of the elevation along each scan line, and thus, the water region always has a higher SLIER value compared to the land. By using this ratio index, one can extract the water surface by looking for those LiDAR data points having high SLIER values. SLIER can overcome the drawback of GMM fitted histogram to split the land and water regions, where SLIER works well on river, coastal region and challenging scenario, such as shore region with land depression.

6. Automatic land-water classification using airborne LiDAR data point clouds

I propose and develop a comprehensive LiDAR data processing workflow to classify the monochromatic/multispectral LiDAR data point cloud for different land-water scenarios, including natural shore, man-made shore, shore with land depression and inland river. I use a new LiDAR-based feature index, SLIER, which can automatically detect the water surface if a scan line completely falls within the water surface surveyed by a linear mode LiDAR scanner. SLIER can be implemented by any monochromatic or multispectral LiDAR data, which are collected LiDAR sensors operated with either oscillating mirror or rotating prism. The experimental work demonstrates that the multispectral LiDAR data is capable of enhancing the classification capability, where an overall accuracy better than 96% was achieved in most of the cases. A ISPRS journal paper is published, and technology transfer (To Teledyne Optech) is accomplished.

7. Scan line void filling of airborne LiDAR point clouds for hydro-flattening DEM

Built based on the SLIER, I propose two sets of algorithms for scan line void filling of classified airborne LiDAR point clouds. These algorithms can help fill up those data voids caused by laser dropouts (on the water bodies) found at the swath edges and the close-to-nadir region. With the filled data voids and assigned with a constant elevation representing the water surface, these can help generate a hydro-flattened DEM, which gets rid of those unpleasant triangular facets. The algorithms have been evaluated with five hydrologic and oceanographic environments as mentioned in the USGS LiDAR Base Specification. The findings are reported in the IEEE JSTARS.

8. VPADS: A video-based pavement distress screening system

This is a collaborative research with the Ministry of Transportation Ontario. I propose and develop a low-cost video-based pavement distress screening system (VPADS) for low volume roads/highways. The system analyzes the video image data collected by any consumer-grade camera mounted in the car front. Since the video image data are collected at oblique view, existing crack detection methods cannot be directly implemented. Therefore, we propose to first define the  RoI by robust line fitting of the two-side road markings [Lane detection for RoI: video1, video2], and subsequently apply a multi-scale ridge detection filter and a boundary contour analysis to evaluate the pavement condition within the RoI. The experimental testings on highway 624 and 668  both yielded an accuracy of 80%. A journal publication can be found here.