TalTech Data Repository
Not a member yet
221 research outputs found
Sort by
Kaugteenuste näidisprojektide protsessi analüüs: III vaheraport
No full text<p>This report was prepared within the framework of the procurement contract concluded between Tallinn University of Technology and the Health Insurance Fund. The purpose of the analysis is to describe and analyse the innovation contest organised by the Health Insurance Fund and the related activities, and to assess the effectiveness of the competition in meeting the objectives of developing remote healthcare services. This report summarises the third phase of the contest. <span> </span></p>
Synthetic dataset for warehouse equipment and pallet recognition (Models)
No full text<p>This dataset provides a collection of synthetic images simulating various warehouse scenarios, specifically designed for training and evaluating computer vision models for equipment and pallet recognition tasks. The images feature a diverse range of warehouse environments, including different lighting conditions, object arrangements, and occlusions.</p><p>This dataset was created as part of a project funded by <a href="https://www.etis.ee/Portal/Projects/Display/196065ff-51bc-48c8-9790-62115a68dec3"><strong>AI & ROBOTICS ESTONIA (EDIH)</strong></a></p><p>Due to upload size limits the dataset was split into 4 pieces.</p><ul><li><a href="/records/nmw7y-41a87"><strong>Camera 1</strong></a></li><li><a href="/records/3ytc7-2z453"><strong>Camera 2</strong></a></li><li><strong>Models</strong></li><li><a href="/records/e9wxe-qpv69"><strong>Videos</strong></a></li></ul><p>For an in depth explaination about the dataset creation and model training process see the thesis in <i>Related works</i>.</p><p><strong>Training process:</strong><br>Both models were trained on the same dataset. The training set consised of ~10000 synthetic warehouse images and ~1300 images of people taken from the <a href="https://cocodataset.org/">COCO dataset</a>.</p><p>Models were trained until they showed a plateau in performance.</p><p><strong>Model type: </strong>instance segmentation</p><p><strong>Models trained:</strong></p><ul><li><a href="https://github.com/IDEA-Research/MaskDINO">MaskDino</a></li><li><a href="https://github.com/baaivision/EVA/tree/master/EVA-02/det">EVA-02 det</a></li></ul>
Position-prediction and MEC resource calculation
No full text<p>Yo can access the dataset through the link below:<br><a href="https://livettu-my.sharepoint.com/:f:/g/personal/oselga_taltech_ee/EspiUHJAACNNo2l0f6AC0B0BqjfF8KIAK4L8apbWSB_FFg?e=JGiu18">Position-prediction-main</a></p><p>The folder includes some functions and datasets for position prediction that are used for service placement and migration task. </p><p>The Datasets (Real and synthetic) used for position prediction.</p><ul><li>Data file obtained from one of the RTK measurements campaign, which is used for the position prediction.</li><li>data_col25.csv: Field trials' dataset. This is one of the Excel files that represent the real dataset (from the measurement campaign). The input features are 25 (The features are COG (course over ground), COGD (COG difference), longitude, latitude, and speed. Five previous measurements are stored and used for each of these features).</li><li>The Excel files are some of the different synthetic datasets that are used for position prediction.</li></ul><p>ANNGridSearch.py: An ANN python script (with grid-search) that performs the position prediction.</p><p>ErlangB.m: An example of two Erlang-B calculation functions. That can be used to calculate the Blocking probability.</p><p>Other real and synthetic datasets, obtained using different devices and from various physical locations, are also available. For inquiries, please reach out to the contact person (Osama Elgarhy).</p>
Seasonal dynamics and regional distribution patterns of CO2 and CH4 in the north-eastern Baltic Sea
No full text<p>We evaluate the variability of carbon dioxide and methane in the surface layer of the north-eastern basins of the Baltic Sea in 2018.</p><p>Dataset contains background meteorological information (files ERA5_longterm_monthly.nc and ERA5_2018_hourly.nc), from CTD profiles calculated upper mixed layer depth information (file UML_CTD2018_sub.xlsx) and continuous surface water pCO2 and cCH4 measurements and calculated fluxes (file Continuous surface water measurements_sub.xlsx).<br><br>Published research article available: https://bg.copernicus.org/articles/21/4495/2024/</p>
Baltic Sea Cod Reproductive Volume from Reanalysis
No full text<p><strong>Generated using E.U. Copernicus Marine Service Information;</strong><br>https://doi.org/10.48670/moi-00013</p>
Synthetic dataset for warehouse equipment and pallet recognition (Camera 1)
No full text<p>This dataset provides a collection of synthetic images simulating various warehouse scenarios, specifically designed for training and evaluating computer vision models for equipment and pallet recognition tasks. The images feature a diverse range of warehouse environments, including different lighting conditions, object arrangements, and occlusions.</p><p>This dataset was created as part of a project funded by <a href="https://www.etis.ee/Portal/Projects/Display/196065ff-51bc-48c8-9790-62115a68dec3"><strong>AI & ROBOTICS ESTONIA (EDIH)</strong></a></p><p>Due to upload size limits the dataset was split into 4 pieces.</p><ul><li><strong>Camera 1</strong></li><li><a href="/records/3ytc7-2z453"><strong>Camera 2</strong></a></li><li><a href="/records/c6182-bgd05"><strong>Models</strong></a></li><li><a href="/records/e9wxe-qpv69"><strong>Videos</strong></a></li></ul><p>For an in depth explaination about the dataset creation and model training process see the thesis in <i>Related works</i>.</p><p><strong>Format:</strong> <a href="https://cocodataset.org/#format-data">COCO</a></p><p><strong>Dataset description:</strong><br>Both camera 1 and 2 contain images from the same virtual warehouse.<br>The difference between the two sets in how much the camera position is randomized between frames.<br>Camera 1 has a wide movement and rotation range to capture objects of interest from as many different angles as possible.<br>Camera 2 mimics a security camera - staying high up and looking down. Its position and orientation is randomized somewhat but it is much more constrained.</p><p><strong>Split:</strong></p><ul><li>Train - 2236 (65%)</li><li>Test - 686 (20%)</li><li>Val - 514 (15%)<br> </li></ul><p><strong>Total:</strong> 3436</p><p><strong>Classes:</strong></p><ol><li>person</li><li>pallet_jack_empty</li><li>mover_pallet</li><li>forklift</li><li>forklift_full</li><li>pallet_jack</li><li>pallet_jack_full</li><li>stacker</li><li>stacker_full</li><li>forklift_empty</li><li>stacker_empty</li></ol>
Data from: Formation of property gradient in coarse-grained niobium using a wedge tool - experiment and analysis
No full text<p>The processes of severe plastic deformation are an effective approach for creating an ultra-fine grain structures and increasing mechanical properties of bulk materials. However, there are usually size limitations for the processed workpieces. Surface treatment processes typically deform only a surface layer of limited thickness in a workpiece or finished part. </p><p>A novel severe plastic deformation process for coarse-grained niobium is introduced in the current work, which employs a tool with an inclined (wedge) surface for deforming the material by reverse shear scheme. The process increases the intensity of shear deformations and the depth of plastic deformation in the body of the workpiece when a wedge tool acts on a surface. The essence of the process is the repeated displacement of the workpiece material in opposite directions during the asymmetrical introduction of a wedge tool until the required degree of deformation is accumulated in the tool-affected volume. This deformation scheme applies a 15° angle wedge tool to a workpiece of 21 mm height. After 9 cycles of plastic deformation, it enabled to create a gradient of the accumulated degree of deformation in the range of true strain e = 0.3 – 4.5. At maximum deformation, the microhardness of the workpieces increased by 1.86 times and the tensile strength by 1.6 times. Fractograms show a significant influence of the accumulated degree of deformation on the nature of fracture. The finite element method simulation of the deformation process showed that creating a uniformly strengthened layer requires at least five deforming operations.</p><p>The proposed reverse shear process with a wedge tool can be used, for example, for improving the structure of the surface layers of niobium ingots for subsequent forming. Due to the creation of a significant gradient of properties, the reverse shear process can be used as an express method for determining the mechanical characteristics of different materials in a wide range of accumulated degree of deformation. </p><p>This upload contains measurement and simulation data, on the analysis of which the manuscript is based. Here are also the images that were used for the figures as well as the photos to document the experiments.</p><p> </p>
Forklift Positioning and Inertial Data Collected in the Industrial Environment
No full text<p>Dataset of the positioning and inertial data, collected with the use of the forklift, naturally operating in the industrial environment. Dataset is provided in the Dataset.csv file and has the following structure: <strong>X, Y, TIME.xy, dt.xy, GYR_X, GYR_Y, GYR_Z, ACC_X, ACC_Y, ACC_Z, MAG_X, MAG_Y, MAG_Z, RLL, PCH, YAW, MAG.YAW, TIME.IMU, dt.IMU, TRU, KEY, MOV</strong>.</p><p>Positioning data was collected with the use of the UWB RTLS system, deployed in the industrial area. The UWB tag was attached to the top of the forklift, and operating at a 5Hz update rate. Provided positioning information includes 2D coordinates (<strong>X</strong> & <strong>Y</strong>) in meters, normalized tag internal clock time (<strong>TIME.xy</strong>) in seconds, and corresponding delta time (<strong>dt.xy</strong>) in seconds. </p><p>Inertial data was collected with the use of 9-DOF IMU unit (Bosh bno055), attached to the top of the forklift. Inertial data was collected at the update rate of 100Hz, and includes triaxial gyroscope readings (<strong>GYR_X</strong>, <strong>GYR_Y</strong> & <strong>GYR_Z</strong>) in degrees per second, triaxial accelerometer readings (<strong>ACC_X</strong>, <strong>ACC_Y</strong> & <strong>ACC_Z</strong>) in meters per second squared, and triaxial magnetometer readings (<strong>MAG_X</strong>, <strong>MAG_Y</strong> & <strong>MAG_Z</strong>) in microteslas. The dataset also includes Euler roll, pitch, and yaw angles (<strong>RLL</strong>, <strong>PCH</strong> & <strong>YAW</strong>) in degrees, calculated by the IMU unit internally, and magnetometer data-based heading (<strong>MAG.YAW</strong>) in degrees, calculated separately for the used local coordinates system. Magnetometer data-based heading includes +90 degrees correction, as the deployed UWB local coordinates system in the test area is rotated -90 degrees in relation to the magnetic north. IMU internal clock time (<strong>TIME.IMU</strong>) in seconds, and the corresponding delta time (<strong>dt.IMU</strong>) in seconds are provided.</p><p>Extra information, provided in the dataset, includes the expected true heading of the forklift (<strong>TRU</strong>) in degrees, and manually recreated using available positioning data and other information from the area, available at the 10Hz rate. A total of 8 designators (<strong>KEY</strong>) are available, indicating moments of time when the forklift was interacting with the payload in the industrial area (picking up or dropping down). The moving or stationary status of the forklift is additionally labeled in this dataset (<strong>MOV</strong>).</p><p>Data from both UWB tag and IMU was collected simultaneously by the same processing unit, providing sensors' synchronization. </p>