此项目使用廉价设备模拟了工业中的焊缝追踪方法,焊缝追踪能够代替人手工解决沉重的或者不方便的零件的焊接问题,此项目在设计的过程中考虑了激光点失去交点,干扰因素的影响等鲁棒性,涉及到的2D和3D技术较为全面
This project serves as a technical demonstration for a CV Algorithm Engineer position, replicating the core pipeline of industrial-grade laser stripe scanners. While utilizing consumer-grade hardware, the system implements a complete suite of anti-interference strategies and sub-pixel precision validation at the algorithmic level. It represents my understanding of robustness, real-time performance, and cost-effective mass-production solutions in industrial vision. Beyond demonstrating the practical application of Computer Vision in factory automation, this project reflects an engineering mindset: pursuing ultimate robustness under limited hardware budgets.
基于单目线结构光三角测量模型，将像素偏移量映射为物理深度：
Based on the laser triangulation model, mapping pixel displacement to physical depth:
对于 V 型/L 型焊缝，激光线的交点即为焊缝位置。我们利用灰度重心法提取亚像素中心线，并通过算法求取交点。结合卡尔曼滤波 (Kalman Filter)，显著改善了系统的鲁棒性，有效避免了焊缝点的跳动或环境干扰。
实现平衡实时性与精度的核心。
** 为了在嵌入式端（如树莓派）实现高帧率，本项目采用 灰度重心法 (Gray-Center-of-Gravity)。通过对激光条纹横截面的灰度分布进行加权计算，突破像素分辨率限制，实现亚像素级定位。 ** To achieve high frame rates on embedded devices (e.g., Raspberry Pi), this project implements the Gray-Center-of-Gravity method, calculating the intensity-weighted centroid to surpass pixel resolution limits:
解决金属反光导致的伪影问题。
** 激光照射在金属表面会产生反光或光晕。在进行 RANSAC 线段拟合前，系统先对激光中心点坐标进行预筛选，剔除斜率异常的噪点。这一预处理确保了输入拟合模型的数据更加纯净，从而极大提升了焊缝定位的准确性。 ** Laser radiation on metallic surfaces often induces specular reflections or halos. Prior to RANSAC line fitting, the system performs a pre-filtering process on the laser coordinates to exclude outliers with abnormal slopes. This ensures a cleaner dataset for the fitting model, leading to significantly higher precision in seam localization.
提升轨迹的平滑度与抗干扰能力。
** 系统采用 卡尔曼滤波 (Kalman Filter) 对计算出的交点坐标进行平滑处理。该机制有效减少了输出坐标的异常抖动，并增强了系统对抗弧光、飞溅等环境噪声干扰的能力。 ** A Kalman Filter is implemented to smooth the calculated intersection coordinates. This mechanism effectively suppresses stochastic jitter in the output data and enhances the system's resilience against environmental interference (e.g., arc light or welding splashes).
兼顾算法鲁棒性与计算效率。 ** 系统根据焊缝位置实时更新 动态搜索窗格 (ROI)。这不仅大幅降低了图像处理的计算开销，还通过排除无关区域的干扰，进一步提升了算法在复杂背景下的鲁棒性。 ** The system generates a Dynamic ROI based on the real-time seam position. This not only significantly reduces computational overhead but also enhances algorithmic robustness by isolating the search space from peripheral background noise.
weld_vision_system
├─ calibrateCamera.py
├─ data
├─ Images
│ ├─ chessboard
│ ├─ deapthImgs
│ └─ hanfeng
│ └─ output.mp4
├─ main0_usingVideo.py
├─ main1_usingCamera.py
├─ main_methods.py
├─ pip0_getBaseHight.py
├─ pip1_getTriangleRatio.py
├─ pyproject.toml
├─ README.md
├─ requirements.txt
├─ tests
├─ utils.py
| 指标 (Metric) | 表现 (Performance) | 备注 (Notes) |
|---|---|---|
| FPS (Processing Speed) | 15 - 30 Hz | 在消费级 PC/嵌入式设备实测 (On standard PC hardware) |
| Static Error (Absolute) |
|
受限于简易激光器线宽 (Limited by laser line width) |
| Success Rate (Recognition) | > 90% | 能有效应对金属表面反光干扰 (Robust against reflections) |
本项目采用普通廉价的线激光发生器与 USB 摄像头。安装时需注意相机与激光平面的夹角应保持在
- 克隆项目:
git clone https://github.com/用户名/项目名.git - 安装依赖:
pip install -r requirements.txt - 运行程序:
- 视频版:
python usingVideo.py - 摄像头版:
python usingCamera.py(需要自己搭建硬件,关键是你你需要固定住摄像头和激光器,然后二者的夹角在45-60°/Hardware setup is required. The key is to securely fix the camera and the laser line generator, maintaining an inclusion angle between$45^\circ$ and$60^\circ$ .)
- 视频版:
本项目代码仅作为个人简历展示及技术交流使用，未经许可严禁任何形式的商用、转载或二次分发。 All Rights Reserved. The source code in this repository is for personal portfolio demonstration and technical exchange only. No part of this project may be copied, modified, or distributed for commercial purposes without explicit permission.