Bundle adjustment (BA) refines the estimation of three-dimensional (3-D) structures and viewing parameters in multiview 3-D reconstruction. It is crucial to the accuracy of the reconstruction. Generic bundle adjustment (GBA) fails to make full use of a priori knowledge of the structure of the objects. We propose a new constrained BA algorithm for 3-D reconstruction of a cylinder surface. In this algorithm, 3-D point positions are parameterized as a combination of the parameters of the cylinder and their two-dimensional locations on the cylinder surface. A random-sample-consensus-based fitting step is used to initialize the parameters of the cylinder surface from more than nine points. Parameters of cameras, the cylinder, and 3-D point locations are then refined by solving the problem of minimizing the reprojection error. Experiments based on synthesized data and real images are performed. Experiment results with synthesized data show that the proposed algorithm achieved a 50% lower reconstruction error than the GBA. Experiment results with real images show that the proposed algorithm is more accurate than the GBA approach when the number of matched point pairs is limited.
Metal bellows is one of important components for mechanical sealing. Because the seam is very thin, and its spatial position is strictly limited, it is very difficult to observe the welding seam in order to control the variables of welding in real-time, the welding quality can not usually be guaranteed under the traditional technology. A novel approach is proposed, which is able to provide an amplificatory and clear image of the micro-plasma welding seam observed from better perspective, then output the image to computer and process it so that the variables of welding can be controlled in real-time. A special optical sensor has been designed and realized according to the approach. The principle and main technologies of the sensor is described in detail. The results of applying the system in practice approved the approach is successful. The welding quality of the inner circle of metal bellows, as well as the working condition of welder has been improved after the system is applied.
In the process of robot Welding, position of welding seam and welding pool shape is detected by CCD camera for quality control and seam tracking in real-time. It is difficult to always get a clear welding image in some welding methods, such as TIG welding. A novel idea that the exposal time of CCD camera is automatically controlled by arc voltage or arc luminance is proposed to get clear welding image. A set of special device and circuits are added to a common industrial CCD camera in order to flexibly control the CCD to start or close exposal by control of the internal clearing signal of the accumulated charge. Two special vision sensors according to the idea are developed. Their exposal grabbing can be triggered respectively by the arc voltage and the variety of the arc luminance. Two prototypes have been designed and manufactured. Experiments show that they can stably grab clear welding images at appointed moment, which is a basic for the feedback control of automatic welding.
Rocket engine is a hard-core part of aerospace transportation and thrusting system, whose research and development is very important in national defense, aviation and aerospace. A novel vision sensor is developed, which can be used for error detecting in arc length control and seam tracking in precise pulse TIG welding of the extending part of the rocket engine jet tube. The vision sensor has many advantages, such as imaging with high quality, compactness and multiple functions. The optics design, mechanism design and circuit design of the vision sensor have been described in detail. Utilizing the mirror imaging of Tungsten electrode in the weld pool, a novel method is proposed to detect the arc length and seam tracking error of Tungsten electrode to the center line of joint seam from a single weld image. A calculating model of the method is proposed according to the relation of the Tungsten electrode, weld pool, the mirror of Tungsten electrode in weld pool and joint seam. The new methodologies are given to detect the arc length and seam tracking error. Through analyzing the results of the experiments, a system error modifying method based on a linear function is developed to improve the detecting precise of arc length and seam tracking error. Experimental results show that the final precision of the system reaches 0.1 mm in detecting the arc length and the seam tracking error of Tungsten electrode to the center line of joint seam.