The Application of Computer Vision in Part Recognition

Abstract

Industrial part recognition, localization, and sorting are essential components of intelligent manufacturing in the Industry 4.0 era. With the rapid development of deep learning, the YOLO series of detectors has become the mainstream solution for industrial vision due to their balance of real-time performance and accuracy. Recent studies have not only improved YOLO architectures—through lightweight design, attention mechanisms, and optimized loss functions—but also explored hybrid strategies that combine deep learning with classical geometric methods to enhance robustness in complex environments. In parallel, complete pipelines that integrate recognition, localization, and robotic grasping have demonstrated feasibility in real-world production scenarios. Experimental results report high accuracy (mAP exceeding 95%), inference speeds above 30 FPS, and localization precision within the sub-millimeter range. Nonetheless, challenges such as limited domain generalization, difficulty with reflective or occluded parts, and the lack of standardized 3D pose benchmarks remain unresolved. Future progress is expected to be driven by synthetic data and domain adaptation, multi-modal sensing, unified multi-task architectures, and industrial-grade deployment pipelines. This review provides a comprehensive synthesis of state-of-the-art approaches and highlights promising directions for advancing intelligent robotic manufacturing.