Laser Vision Sensor

lastmod 2026-04-17created 2026-03-157 edits

An optical automation sensor that projects a laser beam onto a target and analyzes the reflected light with a camera to measure 3D shape, distance, position, and more with high precision down to 0.1 mm in welding and machining processes.

It replaces the human eye to perform weld seam tracking (Seam Tracking) and joint recognition, contributing to productivity gains and quality control.

About

• The complete guide to vision sensors (Vision Sensor): the eyes of machines

A vision sensor (Vision Sensor) is an automated inspection device that captures images of products using an industrial camera and lighting, and analyzes them with dedicated algorithms to distinguish good products from defective ones. It replaces the human eye to find tiny defects at high speed, measures dimensions or passes coordinates to robots, and stores inspection results as digital data to make quality control more reliable; it is a core smart factory solution that makes 100% full inspection possible.

The trends in 2026 vision technology are 'AI deep learning (Deep Learning)' and '3D inspection'. AI learns on its own to judge, with human-level perception, irregular scratches and stains that were hard to catch with fixed rules (Rule-based), and 3D vision technology using lasers or pattern light measures even volume and height, going beyond the limits of conventional 2D inspection.

Example

Basic output data types

Main data formats

Raw / binary formats

• Proprietary binary — can only be parsed through the manufacturer's SDK (Keyence, Cognex, Sick, etc.)
• .bin / .dat — header + profile array structure

Point cloud formats

• .pcd — the Point Cloud Library (PCL) standard format
• .ply — a general-purpose format with high compatibility across various tools
• .xyz / .csv — text formats in the form X,Y,Z or X,Y,Z,Intensity

Image types

• .tiff (16/32-bit) — mainly used to store height maps. Can retain floating-point precision
• .png (8/16-bit) — storage for visualization purposes

Streaming / communication protocols

• TCP/UDP Raw — real-time profile stream transmission
• GigE Vision + GenICam — standard protocols for industrial cameras
• EtherCAT / Modbus — trigger and result delivery for PLC integration

Features by manufacturer

Welding robot integration

The typical data flow for weld seam tracking (Seam Tracking) is as follows.

실시간 스트림 → TCP/UDP Raw Profile → 호스트 파싱 → X/Z 좌표 추출 → ROS2 토픽 발행

Profile data structure

• 1 profile = N (x, z) pairs + timestamp
• ROS2 message type: used by wrapping in sensor_msgs/LaserScan or sensor_msgs/PointCloud2

See also

• AIWeldingRobot:Basic
• Korean: 포인트 클라우드
• ROS2
• GigE Vision
• Korean: 용접 로봇
• Korean: 레이저 삼각측량 (Laser Triangulation)
• Welding:SurfaceSampleDataGenerator - A method for generating "Surface" virtual data as ply by adding multiple lines along the y axis to a single line of csv data from LVS that has only the (x, z) axes.

Favorite site

• Sensor selection guide: a complete breakdown of vision sensors vs. laser sensors for smart factories | Smartfactoria