3D Reconstruction Study of Motion Fuzzy Coded and Non-coded Targets Based on Iterative Relaxation Method

• Published in: Journal of electrical engineering & technology Vol. 20; no. 5; pp. 3525 - 3536
• Main Author: Shi, Yun
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.07.2025; 대한전기학회
• Subjects: Electrical Engineering; Electrical Machines and Networks; Electronics and Microelectronics; Engineering; Instrumentation; Original Article; Power Electronics; 전기공학; Deep convolutional neural networks (DCNNs); Iterative relaxation; Fuzzy-coded and non-coded targets; 3D reconstruction (3D-R)
• ISSN: 1975-0102; 2093-7423
• DOI: 10.1007/s42835-024-02133-x

Robotics and computer vision researchers are struggling to accurately and robustly recreate 3D models of moving objects. This work provides a novel approach based on iterative relaxation for reconstructing the motion of both fuzzy-coded and non-coded targets. To overcome motion blur and the lack of distinguishing characteristics in circular retro-reflective targets, we use deep convolutional neural networks (DCNNs) for feature extraction and representation learning. To do this, we gather a large dataset of motion-blurred images with circular retro-reflective objects and train DCNN architecture. Through this training, the network learned to discriminate and extract relevant information from targets.After training, the DCNN extracts features from motion images. An iterative relaxation strategy refines the original 3D reconstruction (3D-R) estimations. This approach uses DCNN features and motion temporal coherence to refine estimates constantly. Relaxation assures that the reconstructed 3D motion matches the observed images and enables motion blur uncertainty.The efficiency of the suggested technique in recreating the 3D motion of both fuzzy-coded and non-coded targets is shown by our experimental findings. We demonstrate that our approach outperforms conventional techniques. The suggested technique has the potential for use in computer vision, robotics, and augmented reality.