<p>High-precision six-degree-of-freedom (6-DoF) robotic grasping is crucial for complex tasks such as laboratory automation, yet its success is highly dependent on robust visual perception. To address the challenges of object detection and precise 6D pose estimation for chemical vials, which often exhibit complex appearances and partial occlusions, this paper proposes a high-fidelity visual perception framework that significantly improves recognition accuracy and localization precision. Specifically, an improved detection model, YOLOv8-BFS, is designed based on YOLOv8s. It integrates a C2f-BiFormer hybrid backbone and an SPPFCSPC multi-scale context module to enhance adaptability to diverse vial appearances and partial occlusions. Concurrently, a refined pose estimation algorithm, Gen6D-Op, is proposed, which resolves the inaccurate position estimation problem of the Gen6D algorithm based on a collinearity assumption, thereby improving the localization precision of the grasping point. Experimental validation on a custom chemical vial dataset demonstrates that the proposed YOLOv8-BFS model achieves 93.2% in mAP@0.5, a 6.7 percentage point improvement over the baseline; the Gen6D-Op algorithm reduces the average grasping point localization error to 7.45 mm, a decrease of 74.7%. Finally, the efficacy of the perception framework was validated in real-world robotic grasping experiments, guiding a robotic arm to achieve an average success rate of 95%. The results demonstrate that the proposed framework provides a reliable visual perception foundation for high-precision 6-DoF grasping tasks.</p>