Estimation of accuracy in determining the translational velocity of a video camera using data from optical flow
| Autor: | Andrii Molchanov, Vyacheslav Kortunov, Farhadi Rahman Mohammadi |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
0209 industrial biotechnology
Computer science ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Optical flow Energy Engineering and Power Technology Video camera 02 engineering and technology Industrial and Manufacturing Engineering law.invention 020901 industrial engineering & automation law Management of Technology and Innovation 0202 electrical engineering electronic engineering information engineering Computer vision Electrical and Electronic Engineering Block (data storage) business.industry Applied Mathematics Mechanical Engineering Motion vector Computer Science Applications Weighting Noise Motion field Control and Systems Engineering 020201 artificial intelligence & image processing Standard algorithms Artificial intelligence business |
| Zdroj: | Eastern-European Journal of Enterprise Technologies. 4:37-45 |
| ISSN: | 1729-4061 1729-3774 |
| Popis: | The devised approaches are adapted to the complicated conditions of observation in certain real tasks, and are fully operational in those cases when existing standard algorithms fail to give reliable results. We propose a method for determining dynamic motion parameters based on the algorithm of a dense optical flow using a texture analysis. In order to determine an optical flow, we employed a block mapping method that uses adaptively variable size and adaptive motion vector search strategy with weighting the measurements of image blocks, where each block is matched with a texture indicator. A standard block method for estimating optical flow does not imply the use of weighting of the image blocks. A measure of the image block texturization and, consequently, the reliability of the computed motion vector, is determined on the basis of conditionality number of the information matrix. Based on the calculated optical flow, in order to estimate motion parameters, it is proposed to use the least square method that takes into account noise of the measured data. In this case, the minimization is applied at which a contribution to an error is weighed, greater importance is given to the points where the optical flow speed is larger. This is most useful when the measurement of high speeds is more accurate. The norm that produces the best results depends on the noise properties in the measured optical flow. When estimating parameters of the translational motion velocity of the entire image frame, the proposed method considers textural differences of the underlying surface, as well as noise in the measured data of each image block.We presented simulation results of a UAV motion along different types of the underlying surface and estimated the accuracy of determining translational motion parameters using the optical sensor. Experimental results confirm that the application of a texture analysis when evaluating a motion field improves performance by recruiting a reduced number of vectors, as well as this proves to be more accurate in comparison with traditional block brute-force methods |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|