Published 2018
“…Disparity depth map estimation of stereo
matching algorithm is one of the most active research topics in computer vision.In the field of image processing,many existing stereo
matching algorithms to obtain disparity depth map are
developed and designed with low accuracy.To improve the accuracy of disparity depth map is quite challenging and difficult especially with uncontrolled dynamic environment.The accuracy is affected by many unwanted aspects including random noises,horizontal streaks,low texture,depth map non-edge preserving, occlusion,and depth discontinuities.Thus,this research proposed a new robust method of hybrid stereo
matching algorithm with significant accuracy of computation.The thesis will present in detail the
development,design, and analysis of performance on Multistage Hybrid Median Filter (MHMF).There are two main parts involved in our
developed method which combined in two main stages.Stage 1 consists of the Sum of Absolute Differences (SAD) from Basic Block
Matching (BBM)
algorithm and the part of Scanline Optimization (SO) from Dynamic Programming (DP)
algorithm.While,Stage 2 is the main core of our MHMF as a post-processing step which included segmentation,merging, and hybrid median filtering.The significant feature of the post-processing step is on its ability to handle efficiently the unwanted aspects obtained from the raw disparity depth map on the step of optimization.In order to remove and overcome the challenges unwanted aspects, the proposed MHMF has three stages of filtering process along with the
developed approaches in Stage 2 of MHMF
algorithm.There are two categories of evaluation performed on the obtained disparity depth map: subjective evaluation and objective evaluation.The objective evaluation involves the evaluation on Middlebury Stereo Vision system and evaluation using traditional methods such as Mean Square Errors (MSE),Peak to Signal Noise Ratio (PSNR) and Structural Similarity Index Metric (SSIM).Based on the results of the standard benchmarking datasets from Middlebury,the proposed
algorithm is able to reduce errors of non-occluded and all errors respectively.While,the subjective evaluation is done for datasets captured from MV BLUE FOX camera using human's eyes
perception.Based on the results,the proposed MHMF is able to obtain accurate results, specifically 69% and 71% of non-occluded and all errors for disparity depth map, and it outperformed some of the existing methods in the literature such as BBM and DP
algorithms.…”
Get full text
Get full text
Get full text
Get full text
Thesis
