Depth Map Estimation from stereo devices and time of flight range cameras have been a challenging issues in Computer Vision. Distance Estimations from single-pixel histograms of time of flight sensors are exploited in numerous fields. Beyond the several drawbacks such as degradation caused by strong ambient light, scattered and multi-path possibilities, most of the prediction algorithms could be applied to resolve these problems effectively. As these two different tasks are handled in connection with each other, supervised approaches are considered since they provide more robust results. These results are used to train the model to improve three- dimensional geometry information and against major difficulties such as complicated patterns and objects. These approaches are observed according to their accuracy with help of metrics and get improved their performances. This thesis focuses on the analysis of Time-of-Flight and stereo vision systems for depth map estimation and single-pixel distance prediction. State of art algorithms are compared and implemented with additional strategies which are integrated to minimize the error ratio. The histograms which are obtained from Time of Flight Sensor Simulation are exploited as a dataset for single-pixel distance prediction and after that, NYU Dataset is selected for depth map estimation.

Algorithms for 3D data estimation from single-pixel ToF sensors and stereo vision systems

KARAKAYA, UFUK BARAN
2021/2022

Abstract

Depth Map Estimation from stereo devices and time of flight range cameras have been a challenging issues in Computer Vision. Distance Estimations from single-pixel histograms of time of flight sensors are exploited in numerous fields. Beyond the several drawbacks such as degradation caused by strong ambient light, scattered and multi-path possibilities, most of the prediction algorithms could be applied to resolve these problems effectively. As these two different tasks are handled in connection with each other, supervised approaches are considered since they provide more robust results. These results are used to train the model to improve three- dimensional geometry information and against major difficulties such as complicated patterns and objects. These approaches are observed according to their accuracy with help of metrics and get improved their performances. This thesis focuses on the analysis of Time-of-Flight and stereo vision systems for depth map estimation and single-pixel distance prediction. State of art algorithms are compared and implemented with additional strategies which are integrated to minimize the error ratio. The histograms which are obtained from Time of Flight Sensor Simulation are exploited as a dataset for single-pixel distance prediction and after that, NYU Dataset is selected for depth map estimation.
2021
Algorithms for 3D data estimation from single-pixel ToF sensors and stereo vision systems
Computer Vision
Deep Learning
Signal Processing
File in questo prodotto:
File Dimensione Formato  
ubk_thesis_pdfA.pdf

accesso aperto

Dimensione 10.51 MB
Formato Adobe PDF
10.51 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/29228