Xiaolu Li

K-Plane-Based Classification of Airborne LiDAR Data for Accurate Building Roof Measurement

A new classification method based on the k-plane clustering algorithm is proposed to segment the point cloud of a building roof, which is obtained from an airborne light detection and ranging (LiDAR) instrument. In the operation of laser points clustering, 3-D coordinates of laser points in the point cloud are directly used as clustering objects. Fitting planes of laser points in the clusters are generated from the obtained clustering solution, and intersecting lines of the fitting planes are calculated. Using the intersecting lines, the point cloud of the building roof is then segmented. Since calculation for the clustering objects, i.e., the normal vectors of neighboring planes of the laser points, required in the classification methods based on the fuzzy k-means clustering algorithm is avoided in the proposed method, not only is the complexity of the classification procedure reduced, but also the accuracy of classification result is improved. In addition, in the proposed method, to guarantee the effectiveness of the k-plane algorithm, the initial cluster planes are estimated from the elevation image of building roof in advance before the process of clustering operation. The proposed k-plane-based classification method is validated by using a number of real airborne LiDAR point clouds.

Quantitative Evaluation of Impacts of Random Errors on ALS Accuracy Using Multiple Linear Regression Method

The formulas for error propagation from the random error sources to the three-dimensional (3-D) coordinate errors of laser footprints of an airborne laser scanning (ALS) system were deduced. Based on large-scale sample data, multiple linear regression experiments were carried out to obtain the standardized regression coefficients of the random errors for the 3-D coordinate errors under different flight heights (from 500 to 6000 m above ground level). The standardized regression coefficients were used to quantitatively evaluate the impacts of the random errors and to sort the impacts on the order of importance. The variation of the impacts of the random errors with increasing flight height and tilt angle of the mirror was also analyzed. The experimental results provide an important guide for error budget of various sensors in the ALS system so as to effectively suppress or compensate significant errors and to improve the performance of the ALS system.

Terrain slope estimation within footprint from ICESat/GLAS waveform: model and method

Terrain slope of Greenland is usually calculated from surface elevation on a large scale. Novel physical models were established to estimate the terrain slope within laser footprints on a smaller scale. Based on the time-stamped waveforms of ice, clouds, and land elevation satellite/geoscience laser altimeter system (ICESat/GLAS), the physical models for calculating the slopes were established in two cases: zero attitude angles and non-zero attitude angles, respectively. The expressions of waveform slopes are related not only to the waveform width and the mean pulse delay of the time-stamped waveform, both in nanoseconds, but also to the attitude of the satellite and the laser divergence angle. In order to calculate the waveform width and the mean pulse delay, a non-Gaussian mathematical function was proposed to curve-fit the waveforms. The slopes estimated from the waveforms were compared with the slopes calculated from surface elevations. Results show that the two methods generate almost identical slope estimations for the same terrain. Calculation results also indicate that the method for slope estimation proposed in this paper performs better for extreme sloping terrain than for gentle sloping terrain.

A high success rate full-waveform lidar echo decomposition method

A full-waveform Light detection and ranging (LiDAR) echo decomposition method is proposed in this paper. In this method, the peak points are used to detect the separated echo components, while the inflection points are combined with corresponding peak points to detect the overlapping echo components. The detected echo components are then sorted according to their energies in a descending order. The sorted echo components are one by one added into the decomposition model according to their orders. For each addition, the parameters of all echo components already added into the decomposition model are iteratively renewed. After renewing, the amplitudes and full width at half maximums of the echo components are compared with pre-set thresholds to determine and remove the false echo components. Both simulation and experiment were carried out to evaluate the proposed method. In simulation, 4000 full-waveform echoes with different numbers and parameters of echo components were generated and decomposed using the proposed and three other commonly used methods. Results show that the proposed method is of the highest success rate, 91.43%. In experiment, 9549 Geoscience Laser Altimeter System (GLAS) echoes for Shennongjia forest district in south China were employed as test echoes. The test echoes were first decomposed using the four methods and the decomposition results were also compared with those provided by the National Snow and Ice Data Center. Comparison results show that the determination coefficient () of the proposed method is of the largest mean, 0.6838, and the smallest standard deviation, 0.3588, and the distribution of the number of the echo components decomposed from the GLAS echoes is the most satisfied with the situation of full-waveform echoes from the forest area, implying that the superposition of the echo components decomposed from a full-waveform echo by using the proposed method can best approximate the full-waveform echo.

A new method for building roof segmentation from airborne LiDAR point cloud data

A new method based on the combination of two kinds of clustering algorithms for building roof segmentation from airborne LiDAR (light detection and ranging) point cloud data is proposed. The K-plane algorithm is introduced to classify the laser footprints that cannot be correctly classified by the traditional K-means algorithm. High-precision classification can be obtained by combining the two aforementioned clustering algorithms. Furthermore, to improve the performance of the new segmentation method, a new initialization method is proposed to acquire the number and coordinates of the initial cluster centers for the K-means algorithm. In the proposed initialization method, the geometrical planes of a building roof are estimated from the elevation image of the building roof by using the mathematical morphology and Hough transform techniques. By calculating the number and normal vectors of the estimated geometrical planes, the number and coordinates of the initial cluster centers for the K-means algorithm are obtained. With the aid of the proposed initialization and segmentation methods, the point cloud of the building roof can be rapidly and appropriately classified. The proposed methods are validated by using both simulated and real LiDAR data.

A Proposal to Compensate Platform Attitude Deviation's Impact on Laser Point Cloud From Airborne LiDAR

The attitude deviations of an airborne stabilized platform have significant impact on the distribution and point density of the laser point cloud obtained from airborne LiDAR. On one hand, the attitude deviations can cause the laser point cloud to horizontally shift along the scanning direction, leading to the coverage area deviating from the target terrain and resulting in missing scan of some important topography. On the other hand, the attitude deviations can cause the point density to be nonuniform, further deteriorating the elevation accuracy of the digital surface model (DSM) reconstructed from the laser point cloud. Among the three attitude deviations of the airborne stabilized platform, the roll and pitch deviations have more significant impact than the heading deviation. Thus, it is of practical importance to take appropriate steps to compensate the attitude deviations of the airborne stabilized platform, especially for the roll and pitch deviations. In this paper, firstly, an attitude compensation device was designed to compensate the impact of both the roll and pitch deviations in real time. Then, through numerical simulation and semi-physical simulation experiments, the compensation effectiveness of the device was evaluated. The experimental results show that the device can effectively compensate the roll and pitch deviations. After the compensation of the roll and pitch deviations, offsets of the distribution of the laser point cloud were well corrected, and the elevation accuracy of the reconstructed DSM was improved.

On-the-Fly Extraction of Polyhedral Buildings From Airborne LiDAR Data

This letter presents an on-the-fly method for extracting polyhedral buildings from airborne light detection and ranging (LiDAR) data. By using the gridding method, the planimetric position and elevation of laser footprints (normally treated as points) in the obtained scan line are mapped into a data sequence. Then, discrete stationary wavelet transform is applied to analyze the elevation variation in the sequence. Buildings in the scan line can be obtained from the detail wavelet coefficients of the sequence. Moreover, to improve precision of the extraction, the gradients of grid points in the geometric planes of building roofs along the direction of the scan line are calculated and remedied by using the corresponding gradients acquired from the adjacent scan lines. With the proposed on-the-fly method, polyhedral buildings in the scan area can be accurately extracted from laser points along the scan lines during the scanning process. The new method is validated by using a set of real airborne LiDAR data.

Quantitatively Evaluating Random Attitude Measurement Errors' Impacts on DSM Elevation Accuracy From Airborne Laser Scanning

Impacts of random attitude measurement errors (RAME) made by the global positioning system (GPS)/inertial measurement unit (IMU) integrated system on elevation accuracy of a digital surface model (DSM) reconstructed from an airborne laser scanning (ALS) system were evaluated. By numerical simulation, three common terrain models, i.e., a planar, a rectangular, and a hemispheric terrain model, were established, and the scanning processes of the ALS system for the three terrain models were simulated to analyze the effects of RAME on reconstructed DSMs. Furthermore, a semi-physical simulation experimental setup was constructed to verify the results obtained from the numerical simulation. A physical terrain model was scanned to quantitatively evaluate the impacts of RAME on positioning accuracy of a laser point cloud and elevation accuracy of a reconstructed DSM. Experimental results show that under the condition of experimental “flight” height of 1310 mm and affected by two kinds of RAME with different standard deviations, i.e., 0.01° and 0.1°, the RMS values of elevation error of reconstructed DSMs increase 0.04 and 1.56 mm, respectively, corresponding to 0.015 and 0.595 m at actual flight height of 500 m. Therefore, if the elevation error of a reconstructed DSM caused by RAME is requested to be lower than 1 cm under the condition of flight height of 500 m, the random attitude measurement accuracy of the GPS/IMU integrated system should be higher than 0.01° (1σ) at least.

Estimation of cluster centers on building roof from LiDAR footprints

A new method for the estimation of the cluster centers on the building roof from the LiDAR footprints is proposed. The elevation image of the building roof is obtained by applying gridding method on the data of the point cloud. The distribution areas of edge lines and ridge line on the building roof are respectively extracted through the calculation of morphological gradient of the elevation image and the operation of point detection on the erosion image. By the method of Hough transform, the feature lines of the building roof are obtained from the distribution areas of them. On the basis of number and normal vectors of geometrical planes, which are composed of the feature lines, the number of the clusters and the initial values of the cluster centers on the building roof are obtained. The proposed method is validated by a simulation experiment.

Fuel-Type Identification Using Joint Probability Density Arbiter and Soft-Computing Techniques

This paper presents a new method for fuel-type identification by combining the joint probability density arbiter and soft-computing techniques. Extensive flame features were extracted both in the time and frequency domains from each flame oscillation signal and formed an original feature data vector. Orthogonal and dimension-reduced feature data were obtained by using the principal component analysis technique. In order to identify the fuel type, the joint probability density arbiter and soft-computing models were established for each known fuel type by using the orthogonal features. Then, the joint probability density arbiter model was used to determine whether the type of fuel is new or not, and one of the soft-computing models (either a neural network model or a support vector machine model) was used to identify the fuel type if the fuel was one of the known types. Experiments were carried out on an industrial boiler. Four types of coal were tested, and the average success rates of fuel-type identification were higher than 97% in 20 trials. The experimental results demonstrated that the combination of the joint probability density arbiter and one of the two soft-computing techniques was effective in identifying the fuel types (either new or not).