An adaptive cost aggregation method based on bilateral filter for stereo matching

Stereo matching is one of the most dynamic fields in computer vision. Though its relevant research has already stepped into a mature stage, there are still certain challenges to obtain real-time and high-precision disparity maps from stereo image pairs. To solve this problem, in the last decade, several cost aggregation methods aimed at improving the quality of stereo matching algorithms have been introduced. This work aims to design an efficient aggregation cost computation algorithm and hardware based on bilateral filter for stereo matching. Cost aggregation is one of the popular methods for stereo matching due to its efficiency and effectiveness. Their limitation is high complexity and some error near the contour, which makes them not implement in real-time. Furthermore, the weakness makes them unattractive for many applications which require accurate depth information A new method of cost aggregation based on the bilateral filter has been proposed. Whereas the algorithms were generally divided into four steps: Matching cost, computation, Cost aggregation, Disparity computation, and post-processing. A new method for calculating matching costs is proposed that combines the image gradient-based optimization cost and the census transform-based cost. In the cost aggregation stage, for each pixel, an adaptive model support window was generated across the base. Then, using the generated support window as a guide, a bilateral filter was used to aggregate the matching costs within the window. An approach was used to aggregate the matching costs within the window used to determine the ideal contrast for each pixel. Winner Take It All (WTA) for the aggregate cost. To further detect inaccurate matching results, a variance optimization framework based on multiple constraints was used to further detect inaccurate matching results. PSNR (Peak Signal to Noise Ratio) and MSSIM (Mean Structural Similarity) calculations were performed for images and to faithfully quantify quality of the images with an average error of 5.6% and optimization using a bilateral filter would be 94.6%. The hardware compliant algorithm model has been designed using MATLAB /Simulink software. This work also provides new insights into stereo matching algorithm design and the proposed method is the most successful among all cost aggregation methods.