In the application of vision measurement, the black light-absorbing object is difficult to reflect the structured light from infrared emitter of the RGB-D camera. Therefore, an image recognition algorithm based on reference environment information is proposed to acquire the spatial positioning information of black volutes in the depalletizing system. The hardware system of the depalletizing system is mainly constructed of an upper computer, a six-axis industrial robot, an RGB-D camera and an end adsorption device. Firstly, the horizontal position information of each volute placed on the cardboard is obtained by the depth differences between the cardboard and the volute. Then, the depth information of the volute is obtained by the upper cardboard depth through collecting the position of the end vacuum suction cup triggered by feedback signal from vacuum generator. Secondly, a regional planar hand-eye calibration method is developed to improve the calibration accuracy in two-dimensional coordinates. The regional calibration method divides the robot working area into four regions: upper left, lower left, upper right, and lower right. The transformation matrix of each region is calculated separately. Finally, the depalletizing experiment is conducted on the three types of volutes. It is concluded that the average positioning error of the grasping center point of each volute obtained by our method is 3.795 mm, and its standard deviation is 1.769 mm. The average value of regional planar hand-eye calibration error is 4.044 mm, and its standard deviation is 1.501 mm. Under a stack of materials with dimensions of 1350 mm × 1350 mm × 1500 mm, the maximum error is controlled within 15 mm. Additionally, when combined with the end feedback compensation mechanism, the success rate for grasping all three volutes reaches 100%.

The sphericity error is one of the most important tolerances to evaluate the dimensional accuracy of bearing balls. A continuous dynamic measurement system based on an industrial camera and a Z-shaped transparent track is developed to measure the sphericity error of bearing balls. The Z-shaped track, which is considered for the parameter setting of length, width and inclination, is designed to achieve a pure rolling of the ball around three perpendicular axes. During the whole measurement process, 15 images are dynamically captured by an industrial camera and divided into three groups according to the three mutually perpendicular directions. Then, a sub-pixel edge contour of the bearing ball from each image is extracted by an image-processing step, such as image de-noising, contrast enhancement, region-of-interest extraction and edge detection. Finally, the roundness error and sphericity error, as defined by ISO 12181-1 and ISO 3290-1, are acquired from the extracted contour of each image. Three different diameter bearing balls with a precision grade of G1000, representing a sphericity error of approximately 25 μm, are applied as experimental objects. The two times standard deviation roundness error obtained from the contour of each image are all less than 7 μm after six repeated measurements. In addition, the two measurement results respectively acquired from the proposed method and a commercial projection measuring instrument have good consistency.

Accurate segmentation of densely stacked and weakly textured objects remains a core challenge in robotic depalletizing for industrial applications. To address this, we propose MaskNet, an instance segmentation network tailored for RGB-D input, designed to enhance recognition performance under occlusion and low-texture conditions. Built upon a Vision Transformer backbone, MaskNet adopts a dual-branch architecture for RGB and depth modalities and integrates multi-modal features using an attention-based fusion module. Further, spatial and channel attention mechanisms are employed to refine feature representation and improve instance-level discrimination. The segmentation outputs are used in conjunction with regional depth to optimize the grasping sequence. Experimental evaluations on camshaft depalletizing tasks demonstrate that MaskNet achieves a precision of 0.980, a recall of 0.971, and an F1-score of 0.975, outperforming a YOLO11-based baseline. In an actual scenario, with a self-designed flexible magnetic gripper, the system maintains a maximum grasping error of 9.85 mm and a 98% task success rate across multiple camshaft types. These results validate the effectiveness of MaskNet in enabling fine-grained perception for robotic manipulation in cluttered, real-world scenarios.