LiDAR Intensity Images Research Articles

Assessment of Random Forest and Neural Network for Improving Land Use/ Land Cover Mapping from LIDAR Data and RGB Image: A Case Study of Magaga-El-Menia Governorate, Egypt

The goals of this article are to improve classification of land use/land cover information using LIDAR data and RGB images, as well as to compare the performance of various supervised machine learning classifiers (random forest and neural network) for extracting land use/land cover information. The 3D coordinates are first transferred to a high-resolution raster via interpolation. Height and intensity raster grids are formed. Second, various raster maps - a normalized digital surface model (nDSM), the difference of returns, and the LiDAR intensity image -are combined to create a multi-channel image. Five scenarios with different combinations were created. Finally, on the five separate datasets, several classifications based on random forest and neural network classifiers were performed. The classification findings were subjected to a quantitative accuracy check. A comparison of these five methodologies has been conducted. Following that, morphological operations were used to eliminate noise. The results revealed also that the fourth approach is the best followed by the third approach then the last approach then the second approach followed by the first approach. It was discovered that random forest classification outperforms neural network classification in terms of classification accuracy.

Long-Distance Multi-Vehicle Detection at Night Based on Gm-APD Lidar

Long-distance multi-vehicle detection at night is critical in military operations. Due to insufficient light at night, the visual features of vehicles are difficult to distinguish, and many missed detections occur. This paper proposes a two-level detection method for long-distance nighttime multi-vehicles based on Gm-APD lidar intensity images and point cloud data. The method is divided into two levels. The first level is 2D detection, which enhances the local contrast of the intensity image and improves the brightness of weak and small objects. With the confidence threshold set, the detection result greater than the threshold is reserved as a reliable object, and the detection result less than the threshold is a suspicious object. In the second level of 3D recognition, the suspicious object area from the first level is converted into the corresponding point cloud classification judgment, and the object detection score is obtained through comprehensive judgment. Finally, the object results of the two-level recognition are merged into the final detection result. Experimental results show that the method achieves a detection accuracy of 96.38% and can effectively improve the detection accuracy of multiple vehicles at night, which is better than the current state-of-the-art detection methods.

Online Extrinsic Calibration on LiDAR-Camera System with LiDAR Intensity Attention and Structural Consistency Loss

Extrinsic calibration on a LiDAR-camera system is an essential task for the advanced perception application for the intelligent vehicle. In the offline situation, a calibration object based method can estimate the extrinsic parameters in high precision. However, during the long time application of LiDAR-camera system in the actual scenario, the relative pose of LiDAR and camera has small and accumulated drift, so that the offline calibration result is not accurate. To correct the extrinsic parameter conveniently, we present a deep learning based online extrinsic calibration method in this paper. From Lambertian reflection model, it is found that an object with higher LiDAR intensity has the higher possibility to have salient RGB features. Based on this fact, we present a LiDAR intensity attention based backbone network (LIA-Net) to extract the significant co-observed calibration features from LiDAR data and RGB image. In the later stage of training, the loss of extrinsic parameters changes slowly, causing the risk of vanishing gradient and limiting the training efficiency. To deal with this issue, we present the structural consistency (SC) loss to minimize the difference between projected LiDAR image (i.e., LiDAR depth image, LiDAR intensity image) and its ground truth (GT) LiDAR image. It aims to accurately align the LiDAR point and RGB pixel. With LIA-Net and SC loss, we present the convolution neural network (CNN) based calibration network LIA-SC-Net. Comparison experiments on a KITTI dataset demonstrate that LIA-SC-Net has achieved more accurate calibration results than state-of-the-art learning based methods. The proposed method has both accurate and real-time performance. Ablation studies also show the effectiveness of proposed modules.

Improvement of Airborne LiDAR Intensity Image Content with Shaded nDSM and Assessment of Its Utility in Geospatial Data Generation

Improvement of Airborne LiDAR Intensity Image Content with Shaded nDSM and Assessment of Its Utility in Geospatial Data Generation

CNN-Based Individual Tree Species Classification Using High-Resolution Satellite Imagery and Airborne LiDAR Data

Deep learning (DL) has shown promising performances in various remote sensing applications as a powerful tool. To explore the great potential of DL in improving the accuracy of individual tree species (ITS) classification, four convolutional neural network models (ResNet-18, ResNet-34, ResNet-50, and DenseNet-40) were employed to classify four tree species using the combined high-resolution satellite imagery and airborne LiDAR data. A total of 1503 samples of four tree species, including maple, pine, locust, and spruce, were used in the experiments. When both WorldView-2 and airborne LiDAR data were used, the overall accuracies (OA) obtained by ResNet-18, ResNet-34, ResNet-50, and DenseNet-40 were 90.9%, 89.1%, 89.1%, and 86.9%, respectively. The OA of ResNet-18 was increased by 4.0% and 1.8% compared with random forest (86.7%) and support vector machine (89.1%), respectively. The experimental results demonstrated that the size of input images impacted on the classification accuracy of ResNet-18. It is suggested that the input size of ResNet models can be determined according to the maximum size of all tree crown sample images. The use of LiDAR intensity image was helpful in improving the accuracies of ITS classification and atmospheric correction is unnecessary when both pansharpened WorldView-2 images and airborne LiDAR data were used.

Analysis of deriving control points from LiDAR intensity image for orthophoto production

In aerial photogrammetry need control points and has a reference to the land coordinates where the measurement location is carried out. This control point would be used in aerial photo georeferenced. But in practice it often finds severe field conditions that can affect the process of measuring control points, such a forest, mountains area and even conflict areas, which could make the process of measurement at that location impossible. LiDAR Intensity Image is one product of processing LiDAR data that can be used to produce coordinates of ground control points. So that it can reduce or eliminate the use of coordinates from GCP point measurements in the field of georeferenced in aerial photographs. This research is calculating the accuracy orthophoto produced of control points from LiDAR Intensity Image. The data used are LiDAR data and aerial photography using the Bundle Block Adjustment method. The result is orthophoto which tested for accuracy based on PERKA BIG No. 15 of 2014. The results from the CE90 and LE90 tests for orthophoto that use control points from the Premark Intensity Image are 0.433 m and 0.479 m which means fulfilling the scale 1:2500 class 1 for horizontal and vertical accuracy. For the CE90 and LE90 tests that use control points from a combination of intensity image and GCP points from GPS measurements, produce better accuracy of 0.314 m and 0.301 m so that fulfilling the scale 1:2500 class 1 for horizontal and vertical accuracy.

Advanced Lake Shoreline Extraction Approach by Integration of SAR Image and LIDAR Data

Noise and an abnormal distributed-image histogram is the main challenge of using SAR data. From this point of view, this study’s authors motivated the non-use of user-defined input parameters. To achieve this purpose, a fuzzy approach was proposed to extract shoreline from SENTINEL-1A data. The parameters in the processing of the SENTINEL-1A image were generated automatically with LIDAR-intensity-derived object-based segmentation results. The LIDAR-intensity image was segmented with the Mean-shift method. The corresponding result was used to estimate the input parameters for fuzzy clustering of the SENTINEL-1A image. Fuzzy segmentation was proposed, due to the expected large number of values regarding water and land classes except for the pixels along the shoreline. The memberships for land and water classes were separately computed. In the proposed approach, the results from LIDAR and SENTINEL-1A dataset are promising, with differences below 1 pixel (10 m) by evaluation with the used reference vector data.

IMAGE LABELING FOR LIDAR INTENSITY IMAGE USING K-NN OF FEATURE OBTAINED BY CONVOLUTIONAL NEURAL NETWORK

Abstract. We propose an image labeling method for LIDAR intensity image obtained by Mobile Mapping System (MMS) using K-Nearest Neighbor (KNN) of feature obtained by Convolutional Neural Network (CNN). Image labeling assigns labels (e.g., road, cross-walk and road shoulder) to semantic regions in an image. Since CNN is effective for various image recognition tasks, we try to use the feature of CNN (Caffenet) pre-trained by ImageNet. We use 4,096-dimensional feature at fc7 layer in the Caffenet as the descriptor of a region because the feature at fc7 layer has effective information for object classification. We extract the feature by the Caffenet from regions cropped from images. Since the similarity between features reflects the similarity of contents of regions, we can select top K similar regions cropped from training samples with a test region. Since regions in training images have manually-annotated ground truth labels, we vote the labels attached to top K similar regions to the test region. The class label with the maximum vote is assigned to each pixel in the test image. In experiments, we use 36 LIDAR intensity images with ground truth labels. We divide 36 images into training (28 images) and test sets (8 images). We use class average accuracy and pixel-wise accuracy as evaluation measures. Our method was able to assign the same label as human beings in 97.8% of the pixels in test LIDAR intensity images.

IMAGE LABELING FOR LIDAR INTENSITY IMAGE USING K-NN OF FEATURE OBTAINED BY CONVOLUTIONAL NEURAL NETWORK

We propose an image labeling method for LIDAR intensity image obtained by Mobile Mapping System (MMS) using K-Nearest Neighbor (KNN) of feature obtained by Convolutional Neural Network (CNN). Image labeling assigns labels (e.g., road, cross-walk and road shoulder) to semantic regions in an image. Since CNN is effective for various image recognition tasks, we try to use the feature of CNN (Caffenet) pre-trained by ImageNet. We use 4,096-dimensional feature at fc7 layer in the Caffenet as the descriptor of a region because the feature at fc7 layer has effective information for object classification. We extract the feature by the Caffenet from regions cropped from images. Since the similarity between features reflects the similarity of contents of regions, we can select top K similar regions cropped from training samples with a test region. Since regions in training images have manually-annotated ground truth labels, we vote the labels attached to top K similar regions to the test region. The class label with the maximum vote is assigned to each pixel in the test image. In experiments, we use 36 LIDAR intensity images with ground truth labels. We divide 36 images into training (28 images) and test sets (8 images). We use class average accuracy and pixel-wise accuracy as evaluation measures. Our method was able to assign the same label as human beings in 97.8% of the pixels in test LIDAR intensity images.

Rigorous LiDAR Strip Adjustment with Triangulated Aerial Imagery

Abstract. This paper proposes a POS aided LiDAR strip adjustment method. Firstly, aero-triangulation of the simultaneously obtained aerial images is conducted with a few photogrammetry-specific ground control points. Secondly, LiDAR intensity images are generated from the reflectance signals of laser foot points, and conjugate points are automatically matched between the LiDAR intensity image and the aero-triangulated aerial image. Control points used in LiDAR strip adjustment are derived from these conjugate points. Finally, LiDAR strip adjustment of real data is conducted with the POS aided LiDAR strip adjustment method proposed in this paper, and comparison experiment using three-dimensional similarity transformation method is also performed. The results indicate that the POS aided LiDAR strip adjustment method can significantly correct the planimetric and vertical errors of LiDAR strips. The planimetric correction accuracy is higher than average point distance while the vertical correction accuracy is comparable to that of the result of aero-triangulation. Moreover, the proposed method is obliviously superior to the traditional three-dimensional similarity transformation method.

Performance evaluation of classification trees for building detection from aerial images and LiDAR data: a comparison of classification trees models

This study assesses the performance of three classification trees (CT) models (entropy, gain ratio and gini) for building detection by the fusion of airborne laser scanner data and multispectral aerial images. Data from four study areas with different sensors and scene characteristics were used to assess the performance of the models. The process of performance evaluation is based on four criteria: model validation and testing, classification accuracies, relative importance of input variables, as well as transferability of CT derived from one data set to another. The LiDAR point clouds were filtered to generate a digital terrain model (DTM) based on the orthogonal polynomials, and then a digital surface model (DSM) and the normalized digital surface model (nDSM) were generated. A set of 26 uncorrelated feature attributes were derived from the original aerial images, LiDAR intensity image, DSM and nDSM. Finally, the three CT models were used to classify buildings, trees, roads and ground from aerial images, LiDAR data and the generated attributes, with the most accurate average classifications of 95% being achieved. The entropy splitting algorithm proved to be a preferable algorithm for building detection from aerial images and LiDAR data.

Optimum estimator for SPOT-pan and ALS-lidar intensity image registration via tie points

The characteristics of a Système Pour d’Observation de la Terre (SPOT) panchromatic image of a geographic area such as an airfield are different from those of an airborne laser scanner lidar intensity image of the same area. This study investigates the feasibility of locating tie points shared by the target (airborne lidar-intensity) and search (spaceborne SPOT-pan) images. At an approximate point position, image segmentation and enhancement are employed so that a pair of target and search window images tend to resemble each other. The similarity is then measured by utilizing a least-squares image matching algorithm which involves variance-component estimation. The proposed method has a 3–15% higher success rate than an equal-weight matching method. Coordinate differences of the conjugate image points serve as reference value to be used in the least-squares collocation method for pixel-by-pixel transformation. The image registration accuracy from a SPOT-pan to a lidar intensity image is estimated to be ±1.0 pixel.