A method for creating a depth map based on a three-level fuzzy model. (January 2023)

• Record Type: Journal Article
• Authors: Bobyr, Maxim; Arkhipov, Alexander; Emelyanov, Sergey; Milostnaya, Natalya
• Is Part Of: Engineering applications of artificial intelligence. Volume 117:Part B(2023)
• Publication Date: 2023-01
• Subjects: Fuzzy logic -- Depth map -- Stereo vision -- Canny detector -- Fuzzy description logics; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2022.105629
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: A new fuzzy method for creating a depth map is presented in the article. It is based on a combination of Canny detector with a three-level fuzzy system and is designed to improve the accuracy of depth mapping. The first fuzzy model was developed to eliminate the errors inherent in Canny detector. The article presents examples to illustrate that Canny filter is not very sensitive to changes in the shape of the gradient. The second fuzzy model makes it possible to eliminate white color artifacts that appear due to the presence of identical large areas on a stereopair, for instance, the background. The third level of the fuzzy model refines the values of disparity obtained using the second level of the fuzzy model and the distances to the near left and right edges of the images obtained during four passes over the stereopair. It was found during the study that the following factors affect the accuracy of the resulting depth map: labels of membership functions of input variables of a three-level fuzzy system; the combination of t and s-norms in the compositional rules of fuzzy inference. The calculation of the root mean square error values made it possible to evaluate the proposed fuzzy model in relation to similar models. It has been established that value of Root Mean Square Error is minimal with the combination of MEAN-MEAN in the structure of fuzzy inference. And three-level fuzzy system for creating a depth map proposed in the article has a significant advantage over existing analogs. Graphical abstract: Highlights: An approach to creating a depth map based on a three-level fuzzy model is proposed. The first level is intended to edge detection in the image. The second level is intended to create a depth map based on the modified algorithm of the sum of absolute differences. The third level combines the results obtained at the first two levels. Both, numerical modeling and experiment, prove the effectiveness proposed of MFDoES-TLFS.