Projects

Inovatívne geoinformatické metódy pre monitoring distribúcie a pohybu ľudí

SP2022/107

Various methods for monitoring the presence of people outside their place of residence (known from census surveys and population records) are currently being investigated. These are methods based on surveillance of mobile phones, social networks, camera systems, remote sensing of the Earth. From these methods, 2 complementary areas were selected, namely social networks and remote sensing with very high spatial resolution. Each of the areas addresses population dynamics at a different temporal and spatial scale. Social networks make it possible to monitor mobility at a medium level of spatial resolution, but with a high time resolution, while remote sensing makes it possible to study mobility at a very high spatial resolution, but with a limited time resolution. Twitter was chosen from the social networks due to the availability of data, location, good word processing and the nature of the network. The aim of the research is to evaluate the possibilities of evaluation (dynamic population) using methods of analysis of contributions from the Twitter network and remote sensing with a very high resolution in the urban environment. Evaluate mobility in selected cities such as Prague, London and Madrid. The focus will be on the traffic behavior indicated on social networks and subsequently interpreted population dynamics. With the help of remote sensing, the evaluation of changes over time in the micro-scale and the evaluation of population dynamics in the micro-scale for the area of the center of Ostrava and Prague will take place. The mobility of people through social networks is currently being monitored, especially in the context of the COVID19 pandemic [6, 8-9]. Along with mobility, it is also possible to examine the relative connection of individual city districts to each other or their socio-economic or racial similarity [7-10]. In all cases, it works with geocoded tweets (location folder drawn from tweet metadata). Natural language processing methods are used to evaluate tweet content. These are semantic analyzes, LDA modeling of topics [12]. The method of tracking the evolution of position over time is used to monitor the evolution of their position component over time. These can be methods called "Single-day distance" and "Cross-day distance", or the determination of a normalized mobility index [13]. When working with tweets, it is recommended to set different limit values, e.g., that only users with more than two posts per day are selected, as well as that the distance between the two positions is more than 0.5 km [8]. Another form of pre-processing is the selection of only those tweets that have a higher spatial resolution than the city level or the removal of tweets published automatically, e.g., job offers, track closures [13]. The research will focus on geocoding methods, analysis of spatio-temporal population dynamics, differentiation of topics in relation to mobility in pandemic times. Its important aspect is also the international comparison of approaches and results of spatio-temporal analyzes. The second area studied is the study of population dynamics in the micrometer. Terrestrial CCTV systems have limited space coverage and access to them is limited. The rapidly evolving capabilities of aeronautical and satellite remote sensing systems make it possible to monitor the situation better across the board. The resolution of satellite images is still not enough to be able to reliably detect each person individually. Nevertheless, changes in intensity, texture, color tones in the place of occurrence of people (mainly groups) allow such mapping. [1] proposes an algorithm that is based on local properties that are extracted from the intensity and color bands of a satellite image. On the other hand, there is the detection of objects using deep learning models. These models can provide better accuracy, less time, less complexity, and overall better performance. CNN is one of the most widely used neural networks for object detection. The most powerful CNN object detection algorithms include Region-based Convolution Neural Network (RCNN), Fast RCNN, Faster RCNN, Single Shot Multibox Detector (SSD) and YOLO (You Look Only Once) [2]. Due to the work in a heterogeneous environment (urban development), where the background and detected objects are quite similar, it is appropriate to use one of the mentioned algorithms for detection in such an environment, or to create your own CNN. Faster RCNNs have been used [2], [3] to detect their objects of interest. It is important for neural networks to divide the images into test phase, training and validation. Given the use of the CNN architecture, a sufficiently large amount of data will be required, which is very difficult given the cost of the commercial satellite images that will be used for processing [2] - [5]. WorldView -3 and WorldView -4 satellite imagery will be used for the best resolution that commercial resources can provide. Therefore, an important step in the processing of the so-called data enlargement by various techniques (scaling, rotation, horizontal flipping, etc.). Another possibility for CNN detection is to generate your own CNN as in [4]. In all cases, the accuracy of detection using the above techniques is at a very high level, at 78% to 94% [2] - [4]. The Python programming language will be used for processing. The advantage of this language is that it works with many libraries that are recommended for deep learning. Such libraries include TensorFlow, Keras, Caffe, Torch or OpenCV, which will be used [5]. Detection of people from satellite data is not yet much addressed in the literature either. Most of the objects of interest are cars , buildings or trees or animals [1]. The synthesis of knowledge from both parts of the research will allow to evaluate the advantages and disadvantages of individual approaches and assess the possibilities of integration into complex urban information systems, representing the smart city framework. Such importance is important e.g. for crisis management, police forces, security analyzes and solutions, transport tasks, monitoring human mobility and other socio - economic tasks [11]. The project is very important for gaining knowledge about methods of evaluating human mobility, processing big data and streams of crowdsourced geodata, the use of neural networks for population detection from satellite images. The responsible solver of the project is Ing. Martin Zajac, who deals with the issues of data processing from social networks, geoparsing data from social networks and natural language processing, which are directly related to the project. The topic of his DiP is "Geoparsing and the spatial context of social media contributions". As part of his previous bachelor's and master's studies, he completed courses on geodata processing, spatial data analysis and spatial statistics, and addressed this issue during his dissertation on "Spatial analysis of data from social networks in relation to public transport". In the past, he participated in the SGS project aimed at detecting population movement using very high-resolution images. Supervisor doc. Dr. Ing. Jiří Horák also deals with the issue of crowdsourced geodata processing and geoparsing as well as remote sensing, specifically object-oriented classification, and the use of artificial intelligence methods. The research team includes another internal doctoral student and 2 students of the follow-up master's study and their roles in the project correspond to the focus of their dissertations, resp. diploma theses. Ing. Peter Golej deals with the analysis of remote sensing data using advanced methods of processing such data. He also deals with data analysis using neural networks, mainly CNN. Both methods of remote sensing data analysis are directly related to the solved project. The topic of his DiP is "Monitoring of people and traffic flows based on satellite observations"; Bc. Marek Ilenčík - deals with machine learning. The topic of his diploma thesis is "The use of Street View and Satellite Images for estimating real estate prices." He has a good knowledge of python in database tools and has a high level of knowledge of the SQL database language and its spatial extension. ; Bc. Kateřina Rutová - In her bachelor's degree, she dealt with remote sensing. She is currently working on issues of visualization and 3D visualization. The topic of her diploma thesis is "Possibilities of 3D visualization in GIS for the needs of the „Pátrač“ application". She also has good knowledge of working with GIS software, spatial databases and programming for GIS needs. References: [1] B. Sirmacek and P. Reinartz, "AUTOMATIC CROWD ANALYSIS FROM VERY HIGH RESOLUTION SATELLITE IMAGES-Web of Science Core Collection". Https://www.webofscience.com/wos/woscc/full-record/WOS:000358311000029 (seen Dec 08, 2021). [2] S. Rawat, "Airplanes Detection for Satellite using Faster RCNN", Medium, Dec. 09, 2019. https://towardsdatascience.com/airplanes-detection-for-satellite-using-faster-rcnn-d307d58353f1 (see Dec 08, 2021). [3] I. Duporge and O. Isupova, "Using very ‐ high ‐ resolution satellite imagery and deep learning to detect and count African elephants in heterogeneous landscapes - Duporge - 2021 - Remote Sensing in Ecology and Conservation - Wiley Online Library". Https : //zslpublications.onlinelibrary.wiley.com/doi/full/10.1002/rse2.195 (see Dec 08, 2021). [4] D. S, S. Kumar, and DS L, "Detection and Classification of Objects in Satellite Images using Custom CNN", Int. J. Eng. Res. Technol., Vol. 10, No. 6, Jun. 2021, Viewed: Dec 15, 2021. [Online] Available from: https://www.ijert.org/research/detection-and-classification-of-objects-in-satellite-images-using-custom-cnn -IJERTV10IS060287.pdf, https://www.ijert.org/detection-and-classification-of-objects-in-satellite-images-using-custom-cnn [5] K. Hacıefendioğlu, HB Başağa, and G. Demir, "Automatic detection of earthquake-induced ground failure effects through Faster R-CNN deep learning-based object detection using satellite images", Nat. Hazards, vol. 105, no. 1, pp. 383-403, Jan 2021, doi: 10.1007 / s11069-020-04315-y. [6] Applied Artificial Intelligence Institute (A2I2), Deakin University, Melbourne, VIC, Australia et al., “Geolocated Twitter-Based Population Mobility in Victoria, Australia, during the Staged COVID-19 Restrictions,” 22nd ed. (Critical Care and Resuscitation, December 7, 2020), https://doi.org/10.51893/2020.4.SC1; [7] Cate Heine et al., “Analysis of Mobility Homophily in Stockholm Based on Social Network Data,” ed. Wenjia Zhang, PLOS ONE 16, no. 3 (March 9, 2021): e0247996, https://doi.org/10.1371/journal.pone.0247996; [8] Chengbo Zeng et al., "Spatial-Temporal Relationship Between Population Mobility and COVID-19 Outbreaks in South Carolina: Time Series Forecasting Analysis," Journal of Medical Internet Research 23, no. 4 (April 13, 2021): e27045, https://doi.org/10.2196/27045; [9] Yuqin Jiang, Xiao Huang, and Zhenlong Li, “Spatiotemporal Patterns of Human Mobility and Its Association with Land Use Types during COVID-19 in New York City,” ISPRS International Journal of Geo-Information 10, no. 5 (May 18, 2021): 344, https://doi.org/10.3390/ijgi10050344; [10] Jennifer Candipan et al., “From Residence to Movement: The Nature of Racial Segregation in Everyday Urban Mobility,” Urban Studies 58, no. 15 (November 2021): 3095–3117, https://doi.org/10.1177/0042098020978965; [11] Caitrin Armstrong et al., "Challenges When Identifying Migration from Geo-Located Twitter Data," EPJ Data Science 10, no. 1 (December 2021): 1, https://doi.org/10.1140/epjds/s13688-020-00254-7. [12] Di Wang et al., “Real-Time Traffic Event Detection From Social Media,” ACM Transactions on Internet Technology 18, no. 1 (November 4, 2017): 1–23, https://doi.org/10.1145/3122982. [13] Xiao Huang et al., “Twitter Reveals Human Mobility Dynamics during the COVID-19 Pandemic,” ed. Song Gao, PLOS ONE 15, no. 11 (November 10, 2020): e0241957, https://doi.org/10.1371/journal.pone.0241957. The work schedule can be found in the project documentation folder in the form of a Gantt chart. Budget Requirement: 300,000 Scholarships: 96,000 Responsible researcher: 3000 CZK / month Other researchers: Ing. Peter Golej CZK 3,000 / month as the main co-solver, others CZK 1,000 / month Material costs: 10,000 - purchase of consumables, toners, office supplies and paper DHaNM: 30,000 - Purchase of PC components (laser printer, PC accessories (widescreen monitor with high resolution for effective workflow with high resolution satelite images, RAM – analysys of big data, diks for backups of big data, SSD) Services: 59,000 - purchase of WorldView-3 or WorldView-4 data (25 km2 approx. 11 000 without VAT), Geomatica (6500 without VAT), software (or ENVI – 10 500 without VAT) translation fees, publication in a professional journal. Travel costs: 60,000 - Two active participants in the CSUM conference. Departure to the CSUM 2022 Skiathos conference, Greece. If it is not possible to travel to the conference, the conference will be conducted online. Fare will be transferred to services and to strengthen the purchase of additional satellite data. Conference: VŠB-TUO: 15,000 - costs for the organization of the student scientific conference Gisáček 2022 (gaining an overview of various forms of use of machine learning and experience with them, feedback on the presented project methodology at the national level) Directed by: 30,000 Historical influence on previous projects Ing. Martin Zajac Population motion detection using very high resolution images. Year: 2021 Funds: CZK 335,000 Ing. Martin Zajac manually detected elements from the data model in the project in very high resolution images. Use of Global Navigation Satellite Systems to Support Meteorological Nowcasting Year: 2020 Funds: CZK 193,000 Ing. Martin Zajac worked on the SGS project, where he evaluated the characteristics of the atmosphere. Ing. Peter Golej Title: Population motion detection using very high resolution images. Year: 2021 Funds: CZK 335,000 Publikácie: GOLEJ, P. ORLIKOVA, L. HORAK, J. LINHARTOVA, P. STRUHAR, J. Detection of people and vehicles using very high-resolution satellite images, 21. International Multidisciplinary Scientific GeoConference Surveying, Geology and Mining, Ecology and Management, 2021 – yet not indexed GOLEJ, P. HORAK, J. Comparison of vehicles detection using very high-resolution satellite images, GIS Ostrava 2022 – Earth Observation for Smart City and Smart Region - yet not indexedIng. Peter Golej worked on a project where he dealt with various advanced methods of remote sensing image data processing and also dealt with image analysis using neural networks. Title: Monitoring of terrain changes using radar interferometry and UAV Year: 2019 Funds: CZK 430,000 Publikácie: ŠÁDEK, P. STRUHÁR, J. Volume computation of municipal landfill - Comparing GNSS and UAV. 40th Asian Conference on Remote Sensing, ACRS 2019: "Progress of Remote Sensing Technology for Smart Future". Neuvedeno : Asian Association on Remote Sensing, 2020, s. nestránkováno RAPANT, P. STRUHÁR, J. LAZECKÝ, M. Radar interferometry as a comprehensive tool for monitoring the fault activity in the vicinity of underground gas storage facilities. Remote Sensing, 2020, roč. 12, č. 2, s. 271. ŠÁDEK, P. STRUHÁR, J. The evaluation of water pollution with the help of remote sensing tools. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. Volume 42, Issue 3/W8. Hannover : International Society for Photogrammetry and Remote Sensing, 2019, s. 403-408. Ing. Peter Golej worked on this SGS project, where he dealt with the issue of remote sensing and used the necessary knowledge from this issue to solve partial tasks, namely the editing and processing of satellite images. Bc. Marek Ilenčík Title: Population motion detection using very high resolution images. Year: 2021 Funds: CZK 335,000 Bc. On the project, Marek Ilenčík manually detected elements from the data model in very high resolution images. Bc. Kateřina Rutová No previous work on SGS projects.

2022

2022

Ministerstvo školství, mládeže a tělovýchovy

SGS

Specifický výzkum VŠB-TUO

Zajac Martin Ing.