Projects

Detekcia pohybu obyvateľstva pomocou snímok veľmi vysokého rozlíšenia

SP2021/36

The project will be focused on image analysis of Earth observations, in which the options of vehicles´ detection and people will be examined on high spatial resolution images. At present, some authors deal with vehicles´ detection using spatial resolution images of very high resolution [1]–[3]. Advanced methods of image data remote sensing processing such as segmentation, including edge detection, detection by grey level or detection of thresholding are used [1]–[3]. After the segmentation is completed it is necessary to distinguish segments and correctly identify chosen objects (e.g.: vehicles). This classification is carried out by morphometric space features of objects in question such as width, length or more complex shape features [2], [4]. Also, the method of template matching method or inter frame difference method [4] or more advanced techniques of segmentation such as application of SegNet architecture [1] can be applied. Neural networks, by which it is possible to classify individual vehicles and other objects, are used to make the results more accurate. Training images are used in the process of neural networks learning and the settings are optimized in a way that the results above the target data set are the best. It is neural networks CNN and their architectures LeNet, AlexNet or VGG-16 [1], [4] that frequently supply the results for these applications. This is used above all for vehicles´ detection. The detection of people from such data seems rather challenging and also there are not too many publications about this. [5] dealt with the people detection in which he used especially local spatial features of images. For distinguishing crowds of people the method of KDE (Kernel density estimation) was applied. Manually counted vehicles or people are used as a validation data. Achieved accuracy is between 74 and 83 % [2], [5]. The main objective of the project is to evaluate possibilities to identify the traffic and population density from satellite imagery data of high special resolution for urban area. The results will also be used to review changes caused by coronavirus outbreak comparing the status before the outbreak and during the restrictions connected with COVID 19. The satellite imagery, Worldview 3 and other sensors with high resolution such as QuickBird will be used for processing. The special attention will be paid to the choice of appropriate satellite images, their pre-processing, atmospheric corrections and testing of appropriate algorithms for the detection of vehicles and people. Different methods of segmentation will be tested as an important step before the classification itself. The classification will focus on use of artificial intelligence methods such as FFC neural networks or CNN neural networks and their architecture. First of all, the tests will be carried for chosen areas. Following, neural networks training will take part with the use of data training sets of image parts. The images of examined areas will be processed after the training of used neural networks. The areas determined are the city centre of Prague, Prague 1 and Staroměstského náměstí square and its surroundings, and the city centre of Ostrava, most likely Moravska Ostrava part with high concentration of people and vehicles. Based on the availability of appropriate images it will be possible to proses also images from different areas from other parts of the world. The results will be compared to validation data for a particular place. The data processing will be done in Geomatica and R software. The working team consists of PhD students as well as graduated (Master´s degree) students of Geoinformatics. The practical usage if the results of this project will be discussed with public administration representatives (Ostrava City Authority, Prague Institute of Planning and Development in Prague), in particular with Ing. David Witosz from Moravská Ostrava district and a Mgr. Jiří Čtyroký, Ph.D. from IPR Praha. Ing. Peter Golej is the responsible project owner, who deals with remote sensing and image analysis that are directly linked to the project. The topic of his dissertation is “Human and Transport Flows based on Satellite Earth Observation”. During Ing. Golej´s bachelor and master studies he completed subjects on remote sensing and he also dealt with this topic in his diploma thesis “Using Sentinel-1 Data for Creating a Digital Terrain Model “ Ing. Golej also participated in satellite image processing training in Google Earth Engine and MatLab (Google Earth Engine training and ARTMO in 1/2021). In the past he participated in SGS project focused on terrain change monitoring using radar interferometry. Supervisor doc. Dr. Ing. Jiří Horák is also dealing with remote sensing, in particular object based image analysis and the usage of artificial intelligence methods. In the working team there are also two internal PhD students and four students of Master´s degree and their roles correspond with the specialization of their dissertation or diploma thesis. Ing. Petra Linhartová – neural networks application, deep learning and FCC; Ing. Juraj Struhár – degradation of images caused by natural effects; Bc. Tomáš Bražina – persistent object filtering (buildings types) from images; Bc. Marek Ilenčik – time changes evaluations on images (temporal classification and filtering); Bc. Milan Večeř – segmentation methods of image used for land cover analysis; Bc. Martin Zajac, - possibilities of evaluation of people density based on social networks. References: [1] N. Audebert, B. Le Saux, and S. Lefevre, ‘Segment-before-Detect: Vehicle Detection and Classification through Semantic Segmentation of Aerial Images’, Remote Sensing, vol. 9, no. 4, p. 368, Apr. 2017, doi: 10.3390/rs9040368. [2] L. Eikvil, L. Aurdal, and H. Koren, ‘Classification-based vehicle detection in high-resolution satellite images’, Isprs Journal of Photogrammetry and Remote Sensing, vol. 64, no. 1, pp. 65–72, Jan. 2009, doi: 10.1016/j.isprsjprs.2008.09.005. [3] X. Chen, S. Xiang, C.-L. Liu, and C.-H. Pan, ‘Vehicle Detection in Satellite Images by Hybrid Deep Convolutional Neural Networks’, Ieee Geoscience and Remote Sensing Letters, vol. 11, no. 10, pp. 1797–1801, Oct. 2014, doi: 10.1109/LGRS.2014.2309695. [4] Q. Tan, J. Ling, J. Hu, X. Qin, and J. Hu, ‘Vehicle Detection in High Resolution Satellite Remote Sensing Images Based on Deep Learning’, IEEE Access, vol. 8, pp. 153394–153402, 2020, doi: 10.1109/ACCESS.2020.3017894. [5] B. Sirmacek and P. Reinartz, ‘Feature analysis for detecting people from remotely sensed images’, Journal of Applied Remote Sensing, vol. 7, p. 073594, Jan. 2013, doi: 10.1117/1.JRS.7.073594. The work schedule can be found in the project documentation folder in the form of a Gantt chart. Budget: - Studentship: 108 000 - scholarship for students, responsible project owner 3000 Kč/month, other researchers 1000 Kč/month - Material costs: 15 000 - purchase of consumables, toners, office supplies and paper - Small tangible and intagible assets: 20 000 - components to PC (RAM, graphics card, monitor, disk), external HDD - Services: 85 000 - purchase of data WorldView 3 (1 image with an area of 25 km2 costs 14 000 Kč) or images of other sensors such as QuickBird, translation fees, publication fees in journal, SW Geomatica, which will be used then for work at the Department of Geoinformatics and for scientific purposes - Travel costs: 30 000 - travel expenses for the conference - SGEM GeoConference 2021 - Conference VŠB-TUO: 12 000 - costs for the GIS Ostrava 2021 conference - Burden: 30 000

2021

2021

Ministerstvo školství, mládeže a tělovýchovy

SGS

Specifický výzkum VŠB-TUO

Golej Peter Ing.