Even as the application of machine vision grows into automation and robotics with the lower costs and smaller sizes of cameras, the use of machine vision for the traditional task of performing quality inspection is expanding with the capabilities offered by cameras, including unprecedented number of pixels, frame rates, sensitivity and range of wavelength information. Today’s machine vision cameras can offer up to 29 megapixels of information from a single charge coupled device (CCD) sensor, image transfer rates of thousands of frames per second with high sensitivity complementary metal oxide semiconductor (CMOS) sensors, and increased wavelength selectivity with two, three or four CCD cameras.
The use of 3-CCD cameras over traditional Bayer pattern cameras for high resolution color measurements is well understood from the point of view of the sampling of the colors at each pixel. Figure 1 shows the internal optical arrangement within a 3-CCD camera. Unlike a conventional camera where all the light is focused on a single sensor, the incoming light is split into three parts (red, green and blue) with the use of a prism. Each part is then imaged onto a separate sensor that samples that part of the color spectrum with up to 1620x1220 pixels. With the use of compact prism blocks and rugged housings, these cameras are gaining acceptance as the tools for color measurement without compromise.
For even more wavelength-based discrimination, multi-spectral cameras have been built with the addition of a near-infrared channel. Figure 2a shows a 4-CCD camera optical block where the incoming light from the same optical location in the object space is split into four parts (red, green, blue and near-infrared) and imaged on to separate linescan CCDs. The primary use of these cameras is viewing color and near-infrared based defects for a continuous inspection system. A unique camera combines a prism block with 2-CCDs to offer an economical area scan multi-spectral solution. In this camera, the color data from the Bayer pattern sensor is augmented by the near-infrared data from the other sensor. Figure 2b shows the prism block inside the camera.
Typical applications of multi-spectral cameras have been in inspections for defects not visible with a color camera while still using color information to gauge visual appearance.
We have demonstrated a novel use of 2-CCD cameras with one Bayer RGB sensor and one near-infrared sensor by using the near-infrared image to measure shape and height information of objects. The integration of the two CCDs within the camera and the optical alignment along the same viewing axis allows the software to combine the pixel data with the height information from the near-infrared image without additional image registration steps. This enables the software to process and display 3-D images at the full 30 frames per second capability of the camera.
The inspection of shape is based on the use of a structured near-infrared light projected on the object, which allows the height profile of the object to be measured. Figure shows the key components of the vision system required: a) the 2-CCD camera, b) a strobed ring light and c) a line pattern projector. The 2-CCD camera is GigE complaint and has dual Gigabit Ethernet ports that connect to the PC. The strobed ring light is used to provide the white light illumination needed for the color image.
The near-infrared line pattern projector is a 5 Watt 850 nm LED and projects an 8-line pattern on to the camera’s field of view. With the additional integration of a single component (the line pattern projector) to a conventional 2-CCD camera vision system configuration we can obtain the height information, which enables the shape quality measurement.
The images acquired by the camera are shown in Figure 3: color image (d), near-infrared image (e). The rendered 3-D image based on the color and height data is shown in Figure 3 (f). A simple triangulation algorithm was used to estimate the height from the displacement of the lines shown in Figure 3 (g). A calibration process establishes the height proportional to the displacement measured in the images and this height information is combined with the color information to create the 3-D rendering. Although the height information is not obtained at every pixel location, it allows for a measurement of the overall height and identification of shape deformities. The change in curvature of the projected lines allows for the detection of small height deformities. The green lines in the Figure 3 (f) show the regions where the curvature change is not large and the red lines show the regions where an unacceptable curvature change is detected corresponding to a depression in the package.
Figure 4 shows a demonstration of a sortation system built with this technology. The top row of images corresponds to the packages that were sorted by their color and the bottom row shows packages that were failed by their a) shape, b) length and c) height. The system is capable of inspecting 30 packages per second at the full image capture rate of the camera.
This novel use of the 2-CCD camera technology to obtain shape information along with color images has enabled the demonstration of a vision system that can sort by color and shape of objects. This technique can be extended to stereo imaging and further to be used with grid patterns or random patterns for enhancing the height information extracted from the near-infrared image. The simplicity of the configuration of the system allows for its use in many quality inspection applications based on machine vision. V&S
Tech TipsThe use of machine vision for the traditional task of performing quality inspection is expanding with the capabilities offered by cameras.
These include an unprecedented number of pixels, frame rates, sensitivity and range of wavelength information.
Today’s machine vision cameras can offer up to 29 megapixels of information from a single CCD sensor and image transfer rates of thousands of frames per second with high sensitivity CMOS sensors.