Jingwei Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jingwei Zhu has a total of 51 co-authors. Here are the top co-authors by number of shared publications:
- Uwe Stilla
Technical University of Munich (TUM)
3 shared publications - Yusheng Xu
College of Surveying and Geo‐Informatics Tongji University Shanghai China
3 shared publications - Boris Jutzi
Institute of Photogrammetry and Remote Sensing, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
2 shared publications - Ludwig Hoegner
Department of Geoinformatics, Hochschule München University of Applied Sciences, Munich, Germany
2 shared publications - Manoj Kumar Biswanath
Photogrammetry and Remote Sensing, Munich Center for Machine Learning (MCML), Technical University of Munich, Munich, Germany
2 shared publications - Zhixiang Fang
Wuhan University
2 shared publications
Latest Publications
- Open Access
- 3 July 2026
- Published by Copernicus GmbH in ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XI-2-2026, 53-60
ThermalAssist: Towards Efficient Annotation of Thermal Imagery
Thermal infrared (TIR) imaging provides a unique capability to reveal surface heat-transfer patterns of buildings and supports applications such as energy leakage detection, insulation inspection, and building energy monitoring. However, large-scale TIR image analysis by deep learning is still constrained by the lack of reliable annotations, as TIR images often exhibit blurred textures and weak boundaries, which makes manual labeling inconsistent and time-consuming. To address this challenge, we propose ThermalAssist, a geometry-aware and gradient-enhanced framework designed to assist thermal anomaly labeling in UAV-based TIR images. By combining sparse labeling, dense correspondence, and label propagation, the framework efficiently transfers labels across overlapping views while maintaining geometric and semantic consistency. Experiments on the TBBR (Thermal Bridges on Building Rooftops) dataset demonstrate that ThermalAssist achieves an F1-score of up to 0.87 and a mean IoU of 0.69, effectively helping reduce missing annotations and inconsistent boundaries. Compared with the tracking-based SAMURAI method, our approach shows greater robustness under low-texture and low-overlap conditions. This work establishes a foundation for a broader thermal annotation system and validates an important step toward scalable, reliable, and more intelligent labeling of thermal anomaly imagery.
- 3 July 2026
- Published by Copernicus GmbH in ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XI-2-2026, 53-60
Open AccessThermalAssist: Towards Efficient Annotation of Thermal Imagery
Thermal infrared (TIR) imaging provides a unique capability to reveal surface heat-transfer patterns of buildings and supports applications such as energy leakage detection, insulation inspection, and building energy monitoring. However, large-scale TIR image analysis by deep learning is still constrained by the lack of reliable annotations, as TIR images often exhibit blurred textures and weak boundaries, which makes manual labeling inconsistent and time-consuming. To address this challenge, we propose ThermalAssist, a geometry-aware and gradient-enhanced framework designed to assist thermal anomaly labeling in UAV-based TIR images. By combining sparse labeling, dense correspondence, and label propagation, the framework efficiently transfers labels across overlapping views while maintaining geometric and semantic consistency. Experiments on the TBBR (Thermal Bridges on Building Rooftops) dataset demonstrate that ThermalAssist achieves an F1-score of up to 0.87 and a mean IoU of 0.69, effectively helping reduce missing annotations and inconsistent boundaries. Compared with the tracking-based SAMURAI method, our approach shows greater robustness under low-texture and low-overlap conditions. This work establishes a foundation for a broader thermal annotation system and validates an important step toward scalable, reliable, and more intelligent labeling of thermal anomaly imagery.
- Open Access
- 1 February 2026
- Published by Springer Nature in Urban Informatics
- Vol. 5 (1) , 4
Automatic generation of thermal point clouds from TIR images and LiDAR point clouds
3D thermal models are associated with building inspection and energy efficiency evaluation. Fusing Thermal infrared (TIR) images with MLS (Mobile Laser scanning) point clouds enables the generation of thermal point clouds, which combine detailed geometric data with thermal attributes at each 3D point. RGB images are typically used to reconstruct a 3D point cloud and apply thermal textures to the model. Therefore, the generated thermal point cloud heavily relies on accurate RGB reconstruction and scale estimation. In this contribution, we introduce a novel image-feature alignment method to directly co-register TIR images with MLS point clouds. The intensity images are generated from the point clouds, and corresponding feature points are matched with the TIR images. With the estimated corresponding points, the pose can be calculated, and the thermal textures are projected onto the MLS point clouds for thermal point cloud generation. Our method achieves results comparable to manual labeling with a projection error of RMSE 3.4 pixels, offering an efficient and reliable solution to generate 3D thermal models for building energy evaluations.
- 1 February 2026
- Published by Springer Nature in Urban Informatics
- Vol. 5 (1) , 4
Open AccessAutomatic generation of thermal point clouds from TIR images and LiDAR point clouds
3D thermal models are associated with building inspection and energy efficiency evaluation. Fusing Thermal infrared (TIR) images with MLS (Mobile Laser scanning) point clouds enables the generation of thermal point clouds, which combine detailed geometric data with thermal attributes at each 3D point. RGB images are typically used to reconstruct a 3D point cloud and apply thermal textures to the model. Therefore, the generated thermal point cloud heavily relies on accurate RGB reconstruction and scale estimation. In this contribution, we introduce a novel image-feature alignment method to directly co-register TIR images with MLS point clouds. The intensity images are generated from the point clouds, and corresponding feature points are matched with the TIR images. With the estimated corresponding points, the pose can be calculated, and the thermal textures are projected onto the MLS point clouds for thermal point cloud generation. Our method achieves results comparable to manual labeling with a projection error of RMSE 3.4 pixels, offering an efficient and reliable solution to generate 3D thermal models for building energy evaluations.
- Top Cited PapersOpen Access
- 13 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 810-830
TUM2TWIN: Introducing the large-scale multimodal urban digital twin benchmark dataset
- 13 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 810-830
Top Cited PapersOpen AccessTUM2TWIN: Introducing the large-scale multimodal urban digital twin benchmark dataset
- 12 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-16
Generation of Thermal Point Clouds From Uncalibrated Thermal Infrared Image Sequences and Mobile Laser Scans
Monitoring building efficiency is a hot topic for engineers and researchers. TIR (Thermal Infrared) images describe thermal attributes but require professional knowledge for analysis. Moreover, images can hardly describe the thermal attributes of the buildings in all aspects. Therefore, as-built thermal point clouds become the solution. In this case, geometric distortions of TIR images can not be neglected. To generate the thermal point clouds with TIR image sequences, we propose to use MLS (Mobile Laser Scanning) point clouds as control point sets for calibration and a workflow for the thermal point cloud generation. To begin with, intensity images are generated from the point clouds, and corresponding points are detected from the TIR image and intensity image frame. Moreover, the intrinsic parameters are estimated with corresponding points. Finally, the thermal point clouds are generated with calibrated TIR images and MLS point clouds. The result demonstrates that the automatic method can achieve comparable or even better calibration results than the manual method.
- 12 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-16
Generation of Thermal Point Clouds From Uncalibrated Thermal Infrared Image Sequences and Mobile Laser Scans
Monitoring building efficiency is a hot topic for engineers and researchers. TIR (Thermal Infrared) images describe thermal attributes but require professional knowledge for analysis. Moreover, images can hardly describe the thermal attributes of the buildings in all aspects. Therefore, as-built thermal point clouds become the solution. In this case, geometric distortions of TIR images can not be neglected. To generate the thermal point clouds with TIR image sequences, we propose to use MLS (Mobile Laser Scanning) point clouds as control point sets for calibration and a workflow for the thermal point cloud generation. To begin with, intensity images are generated from the point clouds, and corresponding points are detected from the TIR image and intensity image frame. Moreover, the intrinsic parameters are estimated with corresponding points. Finally, the thermal point clouds are generated with calibrated TIR images and MLS point clouds. The result demonstrates that the automatic method can achieve comparable or even better calibration results than the manual method.
- Open Access
- 9 June 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (11) , 1875
TUM-MLS-2016: An Annotated Mobile LiDAR Dataset of the TUM City Campus for Semantic Point Cloud Interpretation in Urban Areas
In the past decade, a vast amount of strategies, methods, and algorithms have been developed to explore the semantic interpretation of 3D point clouds for extracting desirable information. To assess the performance of the developed algorithms or methods, public standard benchmark datasets should invariably be introduced and used, which serve as an indicator and ruler in the evaluation and comparison. In this work, we introduce and present large-scale Mobile LiDAR point clouds acquired at the city campus of the Technical University of Munich, which have been manually annotated and can be used for the evaluation of related algorithms and methods for semantic point cloud interpretation. We created three datasets from a measurement campaign conducted in April 2016, including a benchmark dataset for semantic labeling, test data for instance segmentation, and test data for annotated single 360 laser scans. These datasets cover an urban area of approximately 1 km long roadways and include more than 40 million annotated points with eight classes of objects labeled. Moreover, experiments were carried out with results from several baseline methods compared and analyzed, revealing the quality of this dataset and its effectiveness when using it for performance evaluation.- 9 June 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (11) , 1875
Open AccessTUM-MLS-2016: An Annotated Mobile LiDAR Dataset of the TUM City Campus for Semantic Point Cloud Interpretation in Urban Areas
In the past decade, a vast amount of strategies, methods, and algorithms have been developed to explore the semantic interpretation of 3D point clouds for extracting desirable information. To assess the performance of the developed algorithms or methods, public standard benchmark datasets should invariably be introduced and used, which serve as an indicator and ruler in the evaluation and comparison. In this work, we introduce and present large-scale Mobile LiDAR point clouds acquired at the city campus of the Technical University of Munich, which have been manually annotated and can be used for the evaluation of related algorithms and methods for semantic point cloud interpretation. We created three datasets from a measurement campaign conducted in April 2016, including a benchmark dataset for semantic labeling, test data for instance segmentation, and test data for annotated single 360 laser scans. These datasets cover an urban area of approximately 1 km long roadways and include more than 40 million annotated points with eight classes of objects labeled. Moreover, experiments were carried out with results from several baseline methods compared and analyzed, revealing the quality of this dataset and its effectiveness when using it for performance evaluation. - Open Access
- 20 February 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (4) , 696
Area-Based Dense Image Matching with Subpixel Accuracy for Remote Sensing Applications: Practical Analysis and Comparative Study
Dense image matching is a crucial step in many image processing tasks. Subpixel accuracy and fractional measurement are commonly pursued, considering the image resolution and application requirement, especially in the field of remote sensing. In this study, we conducted a practical analysis and comparative study on area-based dense image matching with subpixel accuracy for remote sensing applications, with a specific focus on the subpixel capability and robustness. Twelve representative matching algorithms with two types of correlation-based similarity measures and seven types of subpixel methods were selected. The existing matching algorithms were compared and evaluated in a simulated experiment using synthetic image pairs with varying amounts of aliasing and two real applications of attitude jitter detection and disparity estimation. The experimental results indicate that there are two types of systematic errors: displacement-dependent errors, depending on the fractional values of displacement, and displacement-independent errors represented as unexpected wave artifacts in this study. In addition, the strengths and limitations of different matching algorithms on the robustness to these two types of systematic errors were investigated and discussed.
- 20 February 2020
- Published by MDPI AG in Remote Sensing
- Vol. 12 (4) , 696
Open AccessArea-Based Dense Image Matching with Subpixel Accuracy for Remote Sensing Applications: Practical Analysis and Comparative Study
Dense image matching is a crucial step in many image processing tasks. Subpixel accuracy and fractional measurement are commonly pursued, considering the image resolution and application requirement, especially in the field of remote sensing. In this study, we conducted a practical analysis and comparative study on area-based dense image matching with subpixel accuracy for remote sensing applications, with a specific focus on the subpixel capability and robustness. Twelve representative matching algorithms with two types of correlation-based similarity measures and seven types of subpixel methods were selected. The existing matching algorithms were compared and evaluated in a simulated experiment using synthetic image pairs with varying amounts of aliasing and two real applications of attitude jitter detection and disparity estimation. The experimental results indicate that there are two types of systematic errors: displacement-dependent errors, depending on the fractional values of displacement, and displacement-independent errors represented as unexpected wave artifacts in this study. In addition, the strengths and limitations of different matching algorithms on the robustness to these two types of systematic errors were investigated and discussed.
- Open Access
- 11 May 2018
- Published by MDPI AG in ISPRS International Journal of Geo-Information
- Vol. 7 (5) , 183
Geographic Prevalence and Mix of Regional Cuisines in Chinese Cities
Previous research on the geographies of food put a considerable focus on analyzing how different types of food or ingredients are consumed across different places. Little is known, however, about how food culture is manifested through various cooking traditions as well as people’s perceptions over different culinary styles. Using a data set captured from one of the largest online review sites in China (www.dianping.com), this study demonstrates how geo-referenced social review data can be leveraged to better understand the geographic prevalence and mix of regional cuisines in Chinese cities. Based on information of millions of restaurants obtained in selected cities (i.e., provincial capitals and municipalities under direct supervision of the Chinese central government), we first measure by each city the diversity of restaurants that serve regional Chinese cuisines using the Shannon entropy, and analyze how cities with different characteristics are geographically distributed. A hierarchical clustering algorithm is then used to further explore the similarities of consumers’ dining options among these cities. By associating each regional Chinese cuisine to its origin, we then develop a weighted distance measure to quantify the geographic prevalence of each cuisine type. Finally, a popularity index (POPU) is introduced to quantify consumers’ preferences for different regional cuisines. We find that: (1) diversity of restaurants among the cities shows an “east–west” contrast that is in general agreement with the socioeconomic divide in China; (2) most of the cities have their own unique characteristics, which are mainly driven by a large market share of the corresponding local cuisine; (3) there exists great heterogeneity of the geographic prevalence of different Chinese cuisines. In particular, Chuan and Xiang, which are famous for their spicy taste, are widely distributed across the mainland China and (4) among the top-tier restaurants ranked by the consumers in a city, the local cuisine is not usually favored, while other cuisines are favored by consumers in many different cities. This study demonstrates the use of social review data as a cost-effective approach of studying urban gastronomy and its relationship with human activities.
- 11 May 2018
- Published by MDPI AG in ISPRS International Journal of Geo-Information
- Vol. 7 (5) , 183
Open AccessGeographic Prevalence and Mix of Regional Cuisines in Chinese Cities
Previous research on the geographies of food put a considerable focus on analyzing how different types of food or ingredients are consumed across different places. Little is known, however, about how food culture is manifested through various cooking traditions as well as people’s perceptions over different culinary styles. Using a data set captured from one of the largest online review sites in China (www.dianping.com), this study demonstrates how geo-referenced social review data can be leveraged to better understand the geographic prevalence and mix of regional cuisines in Chinese cities. Based on information of millions of restaurants obtained in selected cities (i.e., provincial capitals and municipalities under direct supervision of the Chinese central government), we first measure by each city the diversity of restaurants that serve regional Chinese cuisines using the Shannon entropy, and analyze how cities with different characteristics are geographically distributed. A hierarchical clustering algorithm is then used to further explore the similarities of consumers’ dining options among these cities. By associating each regional Chinese cuisine to its origin, we then develop a weighted distance measure to quantify the geographic prevalence of each cuisine type. Finally, a popularity index (POPU) is introduced to quantify consumers’ preferences for different regional cuisines. We find that: (1) diversity of restaurants among the cities shows an “east–west” contrast that is in general agreement with the socioeconomic divide in China; (2) most of the cities have their own unique characteristics, which are mainly driven by a large market share of the corresponding local cuisine; (3) there exists great heterogeneity of the geographic prevalence of different Chinese cuisines. In particular, Chuan and Xiang, which are famous for their spicy taste, are widely distributed across the mainland China and (4) among the top-tier restaurants ranked by the consumers in a city, the local cuisine is not usually favored, while other cuisines are favored by consumers in many different cities. This study demonstrates the use of social review data as a cost-effective approach of studying urban gastronomy and its relationship with human activities.
- Open Access
- 19 April 2018
- Published by MDPI AG in Sustainability
- Vol. 10 (4) , 1256
Understanding the Dynamics of the Pick-Up and Drop-Off Locations of Taxicabs in the Context of a Subsidy War among E-Hailing Apps
The locations where taxicabs pick up and drop off passengers are crucial to understanding the dynamics of taxi trip demand. Investigating their spatial distribution and derived dynamic features is still a key task in the fields of urban geography and transportation. Such investigations are urgently needed, considering the competition created by new communication technology services, specifically e-hailing apps such as Uber, Didi and Kuaidi. For example, a subsidy war between two e-hailing apps occurred in China in 2014. However, how the pick-up and drop-off locations of taxicabs change during subsidy wars is still an open question. This paper introduces a methodological framework that can be used to derive the pick-up and drop-off dynamics of taxicabs. It also proposes three indexes that can be used to assess the dynamics of the pick-up and drop-off locations at the city and sub-district scales, namely the numbers of daily pick ups and drop offs per taxi, average transfer distance per unit area of weighted mean centers of pick-up and drop-off locations, and degree of dispersion in the spatial distribution of pick-up and drop-off locations. This paper employs data from taxicabs in the city of Shenzhen to uncover the dynamics of their pick-up and drop-off locations. The results show that the methodological framework and the indexes introduced are powerful tools for uncovering the dynamics of the pick-up and drop-off locations of taxicabs in urban environments.
- 19 April 2018
- Published by MDPI AG in Sustainability
- Vol. 10 (4) , 1256
Open AccessUnderstanding the Dynamics of the Pick-Up and Drop-Off Locations of Taxicabs in the Context of a Subsidy War among E-Hailing Apps
The locations where taxicabs pick up and drop off passengers are crucial to understanding the dynamics of taxi trip demand. Investigating their spatial distribution and derived dynamic features is still a key task in the fields of urban geography and transportation. Such investigations are urgently needed, considering the competition created by new communication technology services, specifically e-hailing apps such as Uber, Didi and Kuaidi. For example, a subsidy war between two e-hailing apps occurred in China in 2014. However, how the pick-up and drop-off locations of taxicabs change during subsidy wars is still an open question. This paper introduces a methodological framework that can be used to derive the pick-up and drop-off dynamics of taxicabs. It also proposes three indexes that can be used to assess the dynamics of the pick-up and drop-off locations at the city and sub-district scales, namely the numbers of daily pick ups and drop offs per taxi, average transfer distance per unit area of weighted mean centers of pick-up and drop-off locations, and degree of dispersion in the spatial distribution of pick-up and drop-off locations. This paper employs data from taxicabs in the city of Shenzhen to uncover the dynamics of their pick-up and drop-off locations. The results show that the methodological framework and the indexes introduced are powerful tools for uncovering the dynamics of the pick-up and drop-off locations of taxicabs in urban environments.