The LED display is composed of thousands of LED light-emitting diodes. One pixel of the LED display consists of R, G, and B tri-color diodes to display text, pictures, graphic images, and video animations. The advantage of LED compared to the previous display technology is that the LED display is completely unaffected by the size and can be tens of hundreds of meters in size. The LED display is widely used in the square of the city center, and some conference rooms, LED display The brightness that can be achieved is far from the LCD liquid crystal display.
The huge number of light points on the LED display caused the LED display to be very consistent. According to the LED manufacturer's industrial grading standards, the brightness error of the same batch of LEDs from the factory can reach 20%-40%, the chromaticity error is 5nm, and the brightness difference that the human eye can tolerate is below 5%. The chromaticity difference is within 1-2nm, so the display without correction is very poor, far below the requirements of the human eye, and, with the long-term use of the LED display, the exterior of the LED tube The silicone grease layer will oxidize, so that the overall brightness of the LED will decrease, the color will be yellowish, the display effect will continue to deteriorate, and the phenomenon of flower screen and mosaic will appear.
With the widespread use of LED displays, for some LED displays (such as conference rooms) for high-profile applications, LED display correction technology is urgently needed to improve display performance. LED display correction is to find each lamp point through a series of algorithms, calculate the information of each lamp point, and then calculate their difference, and generate a correction coefficient table to control the display of each lamp on the LED display. State, so that the display of each light tends to be consistent.
LED lighting characteristics
LED (Light Emitted Diode) is a device that emits visible or near-infrared light with high efficiency when a semiconductor pn junction or the like is connected with a forward current. The core component of the light-emitting diode is a wafer composed of a P-type semiconductor and an N-type semiconductor, and a transition layer between the P-type semiconductor and the N-type semiconductor is referred to as a PN junction.
In some PN junctions of semiconductor materials, at the forward voltage, when the minority carriers injected are combined with how many carriers, the excess energy is released as light, thereby directly converting the electrical energy into light energy. The PN junction adds a reverse voltage, and the minority carriers are difficult to inject, so they do not emit light. When the PN junction is at a forward voltage, the semiconductor wafer can emit light of different colors from ultraviolet to infrared, and the intensity of the light is related to the current.
PWM modulation
Changing the LED current changes the brightness of the LED. However, due to the different current characteristics of the R, G, B tri-color tubes, direct adjustment of the current will cause the ratio of the three colors R, G, B to change, which will change the chromaticity while adjusting the brightness, so we do not take the method of adjusting the current. To control the brightness of the LEDs, here we use PWM modulation to control the brightness of the LEDs.
Pulse Width Modulation (PWM) is a widely used technique for controlling the brightness of LEDs. The principle is based on Talbot's law. PWM is different from changing the current to adjust the brightness of the LED. It controls the brightness of the LED by controlling the illumination of the LED. The current does not change and the chromaticity does not deviate. As shown in Figure 1, by the high-speed flashing of the LED light, the human eye can't detect the flashing of the light, only the light is brightened or darkened; for example, we control 1/10 of the LED light in 10s. The time is illuminated, of course, this is a process of high-speed change. The process of 10s will generate thousands of pulses, which is the most comfortable for the eyes. At this time, we get the brightness of the LED light, which is the brightness of the light in 10s. /10.
The advantage of this method is obvious, there is no need to change the current, because the use of current adjustment will change the LED chromaticity, and PWM modulation can be used to adjust the brightness of the LED without changing the chromaticity.
Calibration process and steps
With this understanding of LED lights, we need to design a calibration process for LED screens:
1) Camera selection parameter configuration
2) RGB image acquisition
3) Image processing LED point positioning
4) Data analysis generates correction coefficients
5) Upload coefficient control LED screen
For the collection of LED information, we need to use the camera. We need to select the camera and lens that can capture the most realistic effect on the human eye. Here we need to parameterize and adjust the camera. After getting the camera, we need to use it to carry on the LED screen. For a series of data acquisition, it is necessary to collect images of R, G, and B for images, and each acquisition requires multiple photos to prevent errors;
After the camera collects the photos, the photos are processed through the software system we designed, and the light points in the photos are positioned to obtain the position of each light spot in the image for the next step; Obtain the pixel value of each lamp point through this position information, then perform data statistics, select the target value, calculate the correction coefficient of each lamp point RGB tri-color lamp to form a 3*3 matrix, and finally obtain the coefficient matrix. Uploaded to the LED screen controller, through which the display of the LED screen is controlled. The coefficients are adjusted multiple times through the feedback mechanism to achieve the best correction.
(1) Acquisition of RGB images
When we adjust the parameters of the camera, we can start the next step and start taking photos of the LED screen. Because the LED screen is composed of three color RGB tubes, we need to brightness each tube. Correction, so we need to take 3 kinds of images, which are the colors of the R tube, the G tube, and the B tube when they are lit separately, that is, red, green, blue, and three color images to correct each tube, and then they are The resulting coefficients form a 3*3 matrix for uploading to the LED controller;
In order to reduce the error, we will take multiple shots of the same color image, and average the coefficients they produce to achieve the most realistic correction factor for RGB lamp characteristics, which will produce better correction results, and the acquired image. as shown in picture 2.
(2) Image processing and LED point positioning
After getting the image, we need to perform a series of image preprocessing on the image, bad point culling, original value dimming field and so on.
Bad point culling method: Calculate all the lamp point information in one image collected, find the average value of the iso-point brightness, and then some brightness is less than 30% of the average brightness (the actual test results) The light point of the value is regarded as the LED dead point, and the points at some positions are not corrected.
The method of darkening the field is to use the RGB three colors. Each LED pixel is subtracted from the pixel value of the same position display K (the image captured when the LED is not displayed); for example: the R pixel captured by an LED camera The value is (255 13 6) and the K pixel value is (10 2 3), then the R pixel value after subtracting the dark field should be (245 11 3).
LED point positioning: For the captured image, we observe that the captured image is very characteristic, the light points are arranged neatly, and the distribution is very uniform, the edge is very obvious. At this time, we use edge detection to determine The position information of a light point, using the famous canny operator for edge detection, then we will use a minimum box to frame the detected edge, which represents all the pixels of a light point taken by the camera. The pixel information of the point, we can locate and count all the pixel information of one (RGB) light point according to the position coordinates of this box. Figure 3 shows the pixel information and location information of a green light point.
What we need to do next is to make sure that the order of each box is the same as the order of the lights. Here we need to count the information of the box. If we sort all the boxes, the efficiency is very low, because the number of light points on an LED screen is 100,000 to one million, and the sorting workload is very large. Therefore, we need to divide the entire screen into multiple correction modules, separate the entire screen, and perform a pre-processing correction for each small block, which can greatly reduce the workload.
Because the camera's viewing angle is determined and the number of light points taken by the camera is lower, the fineness is lower, which affects the accuracy of the collected data. The module size we selected during the system design process is a 128*96 pixel lamp. Point module, we need to divide the whole LED display into N modules of this size, and then correct each module, which greatly reduces the amount of data and improves the calibration efficiency; and data is performed on a 128*96 module. During the collection process, due to the reduction of the number of lamp points, the position information of the detected lamp points is more accurate;
After detecting this information, we sort the information in the next step so that the order of the boxes is the same as the order of the light points. This is because when the CANNY operator detects the edge, it is the top left corner of the entire image. The first pixel to start detecting is not the one we want to detect from the first light point, so the stretching of the image and the tilt of the light point in the image will cause inaccurate detection;
We need to rotate the image to ensure that the light points are not tilted. At this point, we will detect the angle of the rectangle by using a minimum frame to detect all the light points, and rotate the image to ensure the correct image. .
Then we can sort, and sorting a 128*96 data block is also a big workload. We also found that the data of the same point is very regular, and its position information changes regularly. We don't need Sorting the entire 8192 lamp points, we only need to sort the lamp points in a row, that is, sorting in 128 points can achieve the effect, because the distance difference in any two rows is relatively large, the position The order of the information is basically disordered between the same row. In order to improve the efficiency of the algorithm, we only need to sort the positions of the lamp points in the same row to achieve a good effect, let the lamp point and position box carry out a A correspondence.
Through this step of processing we get the pixel information of all the light points; next we need to count the obtained information and calculate the correction matrix of each light point.
(3) Data statistics and generation of correction coefficients
The system uses the weighted average method to count the (R, G, B) values ​​of a light point, that is, the sum of all the pixel values ​​in a box of a light point and then divide by the number of points (R, The G, B) value is taken as the pixel value of this lamp point. The above data statistics are performed on the images of the three lamps of R, G, and B respectively to obtain the values ​​of red tubes (RR, GR, BR) on the same lamp point, (RG, GG, BG) of the green tube, and blue. The value of the color tube (RB, GB, BB), and then convert the pixel values ​​of the three colors of RGB to the XYZ space value: (The WeChat cannot edit the operator symbol, the following operation is represented by the picture)
Then, the target value of the three-color lamp is obtained. The purpose of the target value is to make all the lamp points change to a target value through a transformation matrix, so as to achieve the purpose of improving consistency. Therefore, the selection of the target value is very important. The most common method is to select the lowest point as the target value, but this method has serious drawbacks. When you select the lowest brightness point as the target value, the corrected screen brightness will be made. The reduction is very powerful, and the effect that people see is not good. In order to achieve a goal of correction and eye comfort, we have developed a set of methods for better target value selection. Example: Find the target value of the red LED, XR target, YR target, ZR target, and explain with XR target:
Once we have the desired target value, we can calculate the 3*3 coefficient K matrix we need, because we want to convert all the light points to the target value, all get the formula XR*KR=XR target, these values You can form a matrix to transform.
(4) The coefficient is uploaded to the LED controller
From the above we can clearly get the K matrix method. When we find the K matrix, the coefficient in K may be negative or greater than 1. At this time, the FPGA chip in the controller that controls the LED display screen cannot handle negative numbers. For decimals, we need to transform the K matrix to make the FPGA chip process. When the coefficient in K is less than 0, Kii is set to 0. When KII is greater than 1, because we use PWM to adjust the brightness, Therefore, when the coefficient is greater than 1, the LED controller cannot increase the pulse width, so all the coefficients greater than 1 need to be set to 1; and because the coefficients accepted by the FPGA can only be integers, all the FPGAs accept 8-bit data. When we need to multiply each coefficient by a 255 to pass it to the FPGA chip;
Experimental result
This system uses vs2008+opencv2.4.6 to implement the image processing direction algorithm, and the results are statistically calculated. We conducted statistical research on the lamp point data before and after the experiment, as shown in Figure 5 and Figure 6.
Figure 5 shows the distribution of the lamp point data when no calibration is performed. It can be clearly seen that the data is very scattered when uncorrected, and the inconsistency is high. The calculated lamp point non-uniformity value is about 14%. Figure 6 shows the corrected value. The data distribution, the non-uniformity value calculated at this time is about 4.1%. From the experimental results, we can see that the lamp point positioning method based on the texture feature is very good, because the position of the lamp point is very high. The accuracy is so that the data obtained is also very close, which can produce better correction results.
Next research goal
The proposed texture-based lamp point localization algorithm of this system is very accurate for obtaining the lamp point data, but for an LED screen, the number of lamp points on it is on the order of 100,000, and all the correction of one such screen takes a long time. Time, so the next research goal is to improve the efficiency of an algorithm.
My research idea for this system is because, for an LED screen, the image taken by the camera, the distance between each lamp point is very close, perhaps based on this, we do not need to detect the texture of each lamp point, only need Texture detection is performed on a few points in the LED screen with special positions. Based on this same distance, we can directly mark the position of other lamp points, which is very important for LED screens with a number of 100,000-level lamp points. The time to correct such an LED screen will be greatly shortened.
Busbar Marine Terminal Block,Screw Busbar Insulators,Electrical Drum Support Busbar,Battery Support Busbar
Wonke Electric CO.,Ltd. , https://www.wkdq-electric.com