Research on Binocular Vision Pose Tracking and Detection Algorithm Based on Deep Learning

Abstract

AbstractIn this paper, the target is segmented by instances based on deep learning. The real-time tracking and detection method of inter-vessel poses is investigated, combining binocular vision techniques to achieve normal estimation of the target, and finally obtaining the target poses by positional decomposition. The algorithm used in this paper does not need to add markers or known target 3D models, so it has good scene applicability and robustness. The main research includes a single-image deep learning pose detection method to address the problem of adding specific markers for traditional visual pose detection. The detection results are tracked using the IOU Tracker tracking algorithm, and the results show that the accuracy of binocular vision pose detection based on deep learning is significantly improved, and the real-time performance meets the requirements.KeywordsDeep learningVisual pose detectionBinocular visionWave compensation