Application of CCD industrial cameras

Select a matrix-like sample point in a certain part of the lane. When a vehicle passes by, the gray level value of the sample point is different from that of the original road surface. If the statistical value of the subtraction between the two exceeds a certain threshold, it indicates the presence of the vehicle.

This method is to lay out virtual detection lines composed of pixels in the direction of vertical or parallel traffic flow, as shown in Figure 2. It is generally composed of bright spots to facilitate the separation of the pixel depth of the road surface and the detection line. When a vehicle passes the detection line, the gray level value on the line is different from when no vehicle passes the road; if the difference in gray level value is greater than a certain threshold, it means that a vehicle has passed. Since the sample point or detection line detection method only captures part of the pixel data for processing, the amount of data processed is significantly reduced, so the calculation time is shortened a lot. In order to meet the requirements of real-time detection, the image processing system AUTOSCOPE, which is currently used in traffic detection, uses detection lines for processing.

On road sections where vehicles are running simply, good results can be obtained by using sample points or detection lines as a means of vehicle detection; however, in complex intersections, how to lay out sample points or detection lines will be the first problem encountered, because vehicles in intersections not only go straight, but also have turning behaviors, and vehicles may appear at any position.

The vehicle detection method that uses full screen processing can obtain more information, but the amount of data to be processed is also significantly increased. There are two detection methods that belong to this method: background subtraction method and binarization method: the background subtraction method takes an image without a vehicle as the background. When the image containing the vehicle is subtracted from the background image point by point, the part of the vehicle is subtracted, such as the TRIP system. The binarization method cuts the image at a certain threshold. The pixel depth higher than this value becomes 255 (white), and the pixel depth lower than this value becomes 0 (black). This can separate the object from the background.

Both the background subtraction method and the binarization method have many shortcomings. The former requires frequent updating of the background, while the latter has a complicated process. The common disadvantage of both is that when the color of the object is similar to the background, it will face the fate of cutting failure. In addition, it is difficult to establish the threshold value, so the multi-valued cutting method is proposed, but the process becomes increasingly complicated.

Foreign R. Taktak, Rita Cucchiara, Cucchiara and others believe that because the information contained in nighttime images is quite different from that of daytime images, the use of algorithms and detection streams will be quite different. Generally speaking, at night and under low lighting conditions, the only eye-catching visual features are car headlights and their beams, street lights, and highly reflective light patterns (such as zebra crossings). They believe that nighttime images are not suitable for motion detection algorithms.

Due to the difference in traffic composition between domestic and foreign countries, foreign studies only identify large vehicles and small vehicles, while domestic studies are more complicated. However, general research simplifies vehicle types into large vehicles, small vehicles and locomotives, and uses these three categories for identification.

When using detection lines or sample points as a way to identify vehicle types, because the amount of data collected is small, it is not conducive to vehicle type identification, so there are fewer people using this method. For vehicle recognition in daytime images, vehicle characteristics such as appearance and size are usually used as classification criteria. Relevant literature is organized as follows.

"Neural Network", which has developed rapidly in recent years and has been applied to many fields, has also been applied to the identification of vehicle types. In addition, the vehicle license plate can also be used to binarize the number image, and identify and record the license plate number through feature matching. Through database comparison, each number can correspond to a certain vehicle type, which can be used to capture illegal vehicles, vehicle counting, vehicle type identification, origin and destination point investigation, travel time analysis, etc.

License plate recognition technology has become increasingly mature in the country in recent years. Some scholars believe that license plate recognition can be divided into three stages: pre-processing, which binarizes the image and removes noise. Then, for license plate positioning, the connecting component labeling method is used to find the connecting components in the image for analysis, and then determine the license plate position. For character recognition, after character segmentation, the size of the structural components is set according to the text size, and finally the typological method is used to find text features for comparison. Some scholars use other methods, such as searching for the license plate and then using pixel segmentation to cut the characters and then use neural networks to identify the characters; or use the calculation of gray-scale conversion numbers to find possible license plate positions, then segment the characters, and then use stroke analysis to identify the characters; or use the changing characteristics of the gray-scale values ​​in the image to find the license plate location, use vertical projection histograms to segment the characters, and use gray-scale value correlation for recognition.