Productive Robotics
3D Weld Seam Detection
Built computer vision pipelines for automatic weld seam detection using industrial depth cameras — covering both 2D image-based detection and 3D point-cloud-based detection, with the OpenCV ecosystem on the 2D side and Open3D on the 3D side.
Developed the 2D image + depth pipeline using OpenCV: adaptive edge detection that handles varying workpiece materials and lighting without manual threshold tuning, combined with depth-gradient information to detect step changes where two metal plates meet. Detected lines are deprojected from image coordinates into 3D robot coordinates using the camera's intrinsics.
Developed the 3D point cloud pipeline in Open3D: standard preprocessing followed by plane segmentation to isolate workpiece geometry from the fixturing, then a groove-detection step that identifies and clusters candidate seam points and validates their linearity to filter out spurious geometry.
Used standard line-fitting techniques to extract the weld seam path from validated clusters and segment results into distinct line segments for workpieces with multiple seams. The two pipelines provide speed-vs-accuracy tradeoffs — 2D is fast for simple joints, 3D handles complex groove geometries and parts with significant depth variation.
Integrated with ROS as a perception node publishing annotated images, 3D markers for visualization, and point cloud data for downstream processing. Supports both snapshot and live modes. Applied depth-camera post-processing filters to reduce noise on reflective and curved metal surfaces — a key challenge in industrial welding environments where shiny metal creates noisy depth maps.