Research On Laser Vision Sensing Seam Tracking Method Based On Discriminant Model

With the improvement of industrial manufacturing level,welding automation technology has gradually become the mainstream and the research hotspot in the field of welding technology.And weld seam identifying and tracking is the premise of realizing efficient automatic welding.Among the numerous sensing methods for weld seam tracking,laser vision sensing method has the advantages of high detection precision,rich visual information and stable transmission,so it is widely adopted.In the actual welding environment,there are strong light,smoke dust and other interferences,which affect the on-line identification accuracy of the actual position of the weld.How to accurately extract the key weld position information in the welding environment with serious noise interferences has become a technological difficulty nowadays.Therefore,this paper took the laser vision sensing seam tracking system as the subject investigated,and the paper carried out related research work on the weld seam visual tracking method.Firstly,a set of weld seam tracking experiment system based on line laser vision sensing method was established,which main contents are the design of line laser vision sensor,the selection of embedded seam deviation correction device,the selection of actuator.The calibration experiment was carried out based on the established visual sensing system,and the chessboard images were used to correct the distortion of the captured images.The linear calibration method based on the standard measuring block was adopted to obtain the actual distance corresponding to a single pixel of the weld image.Then,the feature points recognition technology of the weld laser stripe image was studied.For laser stripe images,region of interest(ROI)extraction,median filtering and stripe centerline extraction were performed.And a square weighted grayscale center method combined with edge detection was proposed to achieve accurate extraction of the laser stripe centerline.For V-shaped weld,a method of feature points recognition based on intercept analysis method and line fitting method was proposed.For I-shaped weld and lap weld,the second order difference method was adopted to quickly extract the weld feature points.Then,the weld seam feature tracking method based on discriminant model was studied.For V-shaped weld and lap weld,an improved tracking-learning-detection(TLD)algorithm for weld seam feature tracking was studied and the experiment results showed that the improved TLD algorithm can realize the accurate identification of the weld center.The weld seam feature tracking based on the kernel correlation filter(KCF)algorithm was studied.The tracking target in the weld image was determined.A ridge regression classifier was trained and the classifier discriminated the target and the background in the weld image to realize the weld feature tracking.The I-shaped weld was tracked during welding and the experiment results showed that the KCF algorithm could track the weld target effectively.Finally,a weld seam tracking method based on the kernel correlation / Kalman filters algorithm was proposed.The anti-interference updating mechanism of the weld target template was added.The Kalman filtering model of the weld center position was constructed and the state amplification method was adopted to process the colored observation noise.The model was utilized to estimate the optimal seam center position.The I-shaped weld and the lap weld were tracked in the experiment and the results showed that the kernel correlation / Kalman filters algorithm had better performance than the original algorithm and further improved the precision of weld seam tracking.