火电厂热力检测系统设计课设_CCD视觉检测系统设计的难点

随着机器视觉技术的发展，CCD视觉检测系统的应用愈加广泛，但是在某些领域仍是未突破的，现在的CCD视觉检测系统设计上仍然存在着什么设计难点呢？

With the development of machine vision technology, the application of CCD vision detection system is more and more extensive, but it is still not breakthrough in some areas. What design difficulties still exist in the design of CCD vision detection system?

今天小编分5个方面和大家一起探讨：发光的稳定性、工件位置、标定、物体运动的速度及软件的测量精度。

Today, we discuss five aspects together: the stability of luminescence, the position of workpiece, calibration, the speed of object movement and the measurement accuracy of software.

1、打光的稳定性

机器视觉测量对光照的稳定性要求非常高，因为光照的变化容易导致图像边缘位置的变化，光照变化10-20%，测量结果将可能偏差出 1-2个像素，再厉害的软件也解决不了这个问题，因此光照的稳定性极其重要。

1. Stability of lighting

Machine vision measurement requires very high stability of illumination, because the change of illumination easily leads to the change of image edge position. The change of illumination is 10-20%. The result of measurement may deviate from 1-2 pixels. No serious software can solve this problem, so the stability of illumination is extremely important.

2、工件位置的不一致性

不管是怎么检测，首先要做的是找到要检测的产品，但有些客户提供的产品，并不是固定位置的，这可能就需要使用机器手臂/夹具等，使检测物品出现在我们的视觉内，但却并不能确保每次的位置都是毫无偏差的。如果定位不准确，测量结果有时会有较大偏差。

2. Inconsistency of workpiece position

No matter how to detect, the first thing to do is to find the products to be tested, but some customers provide products that are not fixed position. This may require the use of machine arm/fixture, so that the detection items appear in our vision, but it can not ensure that each position is unbiased. If the positioning is not accurate, the measurement results will sometimes have large deviations.

3、标定

一般在高精度测量时需要做标定，一光学畸变标定，二投影畸变的标定，三物像空间的标定。

目前标定算法都基于平面标定，如果产品不是平面的，普通的标定算法是解决不了的，只能依靠一些特殊的算法。

3, calibration

In general, calibration is needed in high precision measurement, such as optical distortion calibration, two-projection distortion calibration and three-object image space calibration.

Current calibration algorithms are based on plane calibration. If the product is not plane, ordinary calibration algorithms can not solve it, only rely on some special algorithms.

4、物体的运动速度

如果被测量的物体不是静止的，而是在运动状态，那么一定要考虑运动模糊对图像精度(模糊像素=物体运动速度*相机曝光时间)，这也不是软件能够解决的。

4. Velocity of object

If the measured object is not stationary, but in the motion state, then we must consider the image accuracy of motion blurring (blurred pixel = object motion speed * camera exposure time), which is not software can solve.

5、软件的测量精度

在测量应用中软件的精度只能按照1/2—1/4个像素考虑，最好按照1/2，而不能向定位应用一样达到1/10-1/30个像素精度，因为测量应用中软件能够从图像上提取的特征点非常少。

5. Measurement accuracy of software

In the measurement application, the accuracy of software can only be considered according to 1/2-1/4 pixels, preferably 1/2, but can not reach 1/10-1/30 pixel accuracy as in the positioning application, because the feature points extracted from the image by software in the measurement application are very few.