英文文献.docx

1、英文文献The Technology of Vehicle IdentificationBased on VideoIntelligent transportation technology overview of the Intelligent Transportation System (Intelligence Transport System, ITS,) is the world forefront of the field of transportation research projects, combining electronic and information techno

2、logy, communications technology, automatic control theory, computer technology and traditional traffic engineering theory and many other the theory of the subject, and used in modern transport management system in order to achieve transport services and intelligent management. The traffic monitoring

3、 system is an important part of intelligent transportation systems, this subsystem is mainly responsible for the collection of road traffic flow parameters, such as traffic volume, speed, models, queuing time and length.Currently, the detection method of the road parameters, ultrasonic detection, in

4、frared detection, induction loop detection and video-based detection. Ultrasonic detection accuracy is not susceptible to the block of vehicles and pedestrians, the detection distance is short; infrared detection by the heat source of the vehicle itself, and anti-noise capability is not strong, and

5、thus the detection accuracy is not high; although to a sense coil detection accuracy relative higher, but the requirements set in the pavement structure, and on the road damage, construction and installation is relatively inconvenient, life is relatively short, easy to damage and other shortcomings.

6、In recent years, computer technology, image processing, artificial intelligence and pattern recognition technology continues to evolve based on video detection method in the detection of traffic flow has been more widely used, relative to other traffic flow detection technology, It has the following

7、 advantages: 1. Video test can detect the traffic scene area; (2) relative to other detection methods, investment, low cost; video sensors and other devices, such as camera, ease of installation and commissioning, and on the road the facility does not produce damage; using video detection technology

8、 can be collected. more traffic flow parameters.The introduction of related technologies and image preprocessing, the background extraction using background subtraction method for vehicle detection, in general, not collected the whole lot of video image processing. Here, the destination according to

9、 the lane, in the whole image in a few areas of interest (also known as virtual coils), and regional real-time processing, detection of the vehicle and traffic flow parameters. Continuous video image to extract a number of frames, and the pixel gray values in the virtual coil frame serial number sto

10、red in order in the array, each pixel in the virtual loop point by point to find the gray value histogram, select the highest number of gray values as the background image pixel gray value. In general, select the most frequently occurring gray value of each pixel as a background image corresponding

11、to the pixel gray value is very reasonable, If you hit a vehicle intensive, can be appropriately increased the number of frames collected, in order to get a more good results. The background extraction requires an initialization process in the extraction process in the background, not to read into t

12、he video image vehicle detection. The background updating from time to time background subtraction and background updating method: set a timer in the program, from time to time, the program began a new round of background extraction in the extraction process in the background, the program vehicles t

13、esting. However, this time with the background for the last time to extract the background, Extract after when the current context, covering the background with the current background image next frame of the vehicle detection using the updated background, the current background for testing. This met

14、hod can effectively suppress the slow changes of light and natural conditions, and to improve the background subtraction algorithm to detect the effect of the vehicle. When the vehicle passes through the virtual coil, virtual coil within the absolute difference between the current grayscale value of

15、 all pixels and the background of the corresponding pixel gray value sum of the change process, and the image intensity of all pixels in the coil the sum of absolute difference between the value of the background image gray value variables. Virtual coil no vehicle passes through the case, the virtua

16、l coil current image information remained relatively constant, small changes of the gray value of pixels within the coil. Then, when the vehicle began to enter the virtual coil region, due to the huge difference between the pixel gray value of background and vehicles will lead to gradually increasin

17、g; vehicles to leave the time value of the virtual loop area will also gradually decrease. When the vehicle left the time value becomes small. This approach may have shortcomings and weaknesses: frame difference is used, three in a row 2 2 differential, although this method has a strong self-adaptiv

18、e, but the difference of successive frames to choose the timing requirements of high , but also depends on the velocity of moving objects, if the movement speed is faster, and the selected time interval is too long, it will result in coverage area between two moving objects, which can not be divided

19、; If the motion is too slow, and the time selected is too small will cause excessive overlap, the worst case, the object is almost completely overlap, there is an object not detected.Computer image filtering two categories: a class method in the spatial domain processing a variety of processing: the

20、 image in the image space; other methods of space images after the changes, such as Fourier-transform,a variety of processing in the frequency domain, and then change back to the image space domain, the formation of the processed image. Frequency domain processing methods Fourier transform and inver

21、se transform, and a variety of wavelet transform and inverse wavelet transform. These methods use the computer memory and computing time is expensive, not suitable for real-time system of the intelligent vehicle. Therefore, the spatial domain median filtering approach, the ways a local average smoot

22、hing technique, pulse interference and impulse noise suppression effect. Under certain conditions, to overcome the linear filter, such as minimum mean square filter and mean filter image detail is blurred, the effective protection of the edges of the image. The statistical characteristics do not nee

23、d an image in actual operation, so this is a great deal of convenience for the image preprocessing. Image preprocessing - edge enhancement based on visual theory shows: identification of an object is from the edge of the beginning of an image different parts of the edge is often the most important f

24、eatures of the pattern recognition. The edge of the surrounding pixel gray scale step change or roof changes the set of pixels, which widely exists between objects, between objects and background, between the primitive and primitive in the image collected by the machine vision system the edge of the

25、 lane information is lost in the background between. The edge enhancement is aimed at highlighting the edge of the road, in order to facilitate road boundary identification.In addition, the edge enhancement algorithm also help overcome the effects of the road uneven illumination. Commonly used in ed

26、ge enhancement operator Robert operator, Sobel operator, Krisch operator, Prewitt operator, the Laplace operator. Here, edge detection using Sobel operator, it is actually a first-order differential operator, which can effectively eliminate most of the useless information in the road image. Discrete

27、-BCDF algorithm is defined as the following formula $ of Sobel operator with a strong ability to suppress the noise, in fact the essence of the Sobel operator to reflect adjacent or some distance away from the pixel gray-scale differences in characteristics. Road in terms of the physical properties

28、close to normal on the road, and the light is generally uniform. In this case, the gray values of neighboring pixels or less after a Sobel operator, put this close to extent into a certain value, the value is usually close to the value. The thesis of the road boundary and other parts of the road wit

29、h a certain gray level differences of Sobel operator to highlight the border pixel value differences, relative to other road part of the value. While the edge of the gradient direction information, and the algorithm is simple, easy to implement. After the image after the Sobel operator operation, bo

30、rder stand out from the whole image. Image preprocessing - Binarization image binarization processing key is reasonable selection of the threshold, the threshold is set too easy to produce noise threshold is set lower resolution of the General Assembly, non-noise signal is considered noise filtered

31、out, taking into account the general test in a more uniform light or strong changes in light conditions, so select the overall optimal threshold method for image binarization processing principle of the overall optimal threshold is a statistical each piece of image gray distribution characteristics,

32、 the use of category variance as a criterion to select the maximum value of between-class variance as the selected threshold binary image using the optimal threshold algorithm to further reduce noise for subsequent Hough transform relatively clean image data, image preprocessing - camera calibration

33、 and set up video cameras can be installed on the overpass on the high-rise building, or a sufficiently high pole. As for the cameras perspective is determined in the calibration time. camera the configuration from the two-dimensional space to three-dimensional image mapping the actual acquisition of three-dimensional image three-dimensional coordinates in the two-dimensional