Modelling of Intelligence Stereoscopic Vision System Using Adaptive Super-Pixel Based Disparity Estimation Algorithm

摘要

One of the areas of computer vision that has typically received the greatest attention is stereo perception of objects. Many contemporary applications, including augmented reality, robotic navigation, and automotive ones, use stereo matching. To solve the shortcomings of conventional key point-based disparity estimating methods is the driving force behind this effort. Traditionally, parallax estimation has been relied on local knowledge of significant locations. However, significant points with the same descriptor might show up in a symmetrical pattern or be sparse in smooth areas. As a result, in smooth symmetric regions, standard key point-based disparity estimating approaches may function only partially as expected. The suggested algorithm is based on super pixels. We use key point and semi-global information to calculate disparity for our suggested approach rather than doing key point matching. The accuracy of disparity estimate is increased by using both key point local information and super pixel semi-global information, especially for smooth and symmetrical regions. For the purpose of extracting depth information from a set of two-colour stereo images, the Fast Fuzzy C Means (FFCM) depth extraction algorithm was created. Edges are preserved while impulse noise and incorrect disparity estimates are removed using an adaptive bilateral filter. Using a collection of testbench photos, we confirm the effectiveness of the suggested strategy. Combining data from linear perspective and texture attributes yields the depth value for the region corresponding to the background. We suggested a methodology that combines vanishing line and foreground segmentation techniques. The viewer can distinguish between the foreground and backdrop using the foreground segmentation algorithm's separate results.