Neural Network-Based Stereo Vision Outlier Removal

Stereo vision systems rely on accurate feature matching to provide valid stereo reconstruction and pose estimation. This accuracy is achieved through outlier removal techniques, such as RANSAC. However, images also contain semantic information, which can be extracted using neural networks. paper proposes an additional method, where the are semantically segmented a network, before features identified assigned identifiers probabilistic data association technique, matches evaluated based this added information. blending of feature-based techniques with dense maps allows for more information tied each feature, not just its position in image. opens paths applications like class-based clustering. The approach proposed compared traditional system by comparing produced disparity values known ground truth measurements, assessed execution speed. It shown how addition segmentation does improve measurements images, loss processing mitigated utilising specialised hardware.