Engineering, etc
(research objectives, research background and present situation, work progress and scheme, etc.). )
BR/>; See appendix.
1, the advanced nature of the project:
With the rapid development of digital information technology and network technology, in the post-PC era, the performance and high performance of embedded processors have been able to meet complex applications such as self-evaluation complex algorithms, and embedded applications will inevitably enter various fields. On the other hand, with the rapid development of China's economy and the hosting of the Beijing Olympic Games, "Intelligent Transportation" will become an indisputable hot topic. Due to the particularity of transportation industry and the strict requirements of technical parameters and use conditions of its equipment, embedded intelligent transportation equipment is an inevitable trend. Because embedded automatic vehicle identification system is an important part of intelligent traffic management system, it is a perfect combination. Embedded technology and vehicle identification technology, including embedded license plate recognition, embedded vehicle identification and automatic color recognition, strive to lock the car at one time.
It has the following advantages:
1, high independence: independence of using embedded technology and connecting with application system only through communication interface.
2. Complete functions: the recognized license plate, logo and color are highly targeted and powerful with the existing system at one time.
3. Plasticity: it can be combined with the built-in wireless network and various serial interfaces at the upstream end of the front-end signal trigger device, which greatly expands the function and application scope of the downstream product system.
4. Convenient maintenance:
2, operability and realizability:
At present, license plate recognition and vehicle recognition, you wait for the maturity and perfection of technology, and the relevant information is easier to obtain. The existing embedded technology is mature, so it is easier to realize the theme than other frontier sciences in terms of technical difficulty. Topics including equipment and materials are also relatively easy to obtain and the cost is moderate.
3. Innovation:
The existing license plate recognition equipment usually uses computers to process data, and some even need the cooperation of several computers, which takes up a lot of space and resources. Even if the embedded system is completed occasionally, its function is limited license plate recognition or car recognition. The systematic and creative combination of embedded license plate recognition, vehicle recognition and automatic color recognition is a one-time solution. The bloated equipment system is difficult to integrate, has poor stability and is difficult to maintain, which is a functional problem.
4. Possible problems:
At present, the main problems are embedded integration and wireless transmission distance. Ideally, we assume that most computers are used to process data, and we can develop a portable wireless data transmission system and an automatic identification system. However, due to the limitation of our time, energy and money, the degree of portability is the biggest problem. In addition, the image recognition of speed and depth of field is a problem that we may face.
Expected result
Specific forms of achievements, such as patents, published papers and physical production technologies (including software programs), can be various forms of achievements.
We look forward to our experimental results.
First of all, we plan to build a complete embedded system technology, which is related to tangible results.
Secondly, we analyze the market situation, and the market prospect is very promising. Embedded vehicle identification system can be patented and produced in the market.
The third aspect, the color of the car, the license plate and the main combination of the car, is to determine the appropriate algorithm, so it is inevitable in the process of system completion, and the design of the algorithm is completed, which is the manifestation of some achievements published in the paper.
Because we plan to complete the system, we need to complete the hardware and software system. From the point of view, a large part of software and algorithm results can be published in newspapers and put into production, and can be reflected by patented hardware results. Undoubtedly, our research results will not only be part of software or only the theme of hardware, which is a great advantage.
Budget content, budget amount and estimated execution time required for the experimental environment.
CCD camera front-end image acquisition, buy a camera or video camera 3000 07. 12 to 08.2.
Light addition of auxiliary light source in special environment 1500 07.5438+0208.2 month.
The analog signal of the image acquisition card was digitized for 250007.438+0208.2 months.
> Hardware facilities of embedded system, image processing from 400008.3 08.438+00 to 08.38+02? 08. 10
DVR video information memory 2500
The image recognition result output by the display device is 1500 08. 1209.2 months.
Wireless transceiver or wired transmission equipment information transmission 250 009 2? 09.3
The final stage of assembling machine parts into 2000 prototype
Total: 19500 yuan
University identity view
Evaluation opinions of the Committee of Experts
School examination and approval opinions
Attachment: the present situation, background and significance of the theme,
From 1885 Since the birth of the first car in the world, cars have brought great influence to our daily work and life. For more than 100 years, cars and their advantages of low cost and convenient operation have been gradually accepted by the public. Every year, hundreds of thousands of families in China join the ranks of car owners. Secondly, the rapid development and convenient lifestyle have caused a series of problems: cars are stolen every year, traffic accidents happen from time to time ... There is no doubt that cars need standardized management. Now, our car management is done by people. It is easy to imagine that there is nothing you can do in the face of the growing fleet manual. Therefore, intelligent transportation will become the inevitable trend of traffic management development in the future.
You can't be sure that traffic intelligence will automatically identify itself. As early as the 1990s, automobile signs attracted worldwide attention, and people began to study the problems related to automatic identification cards-automatic identification of automobile license plates. A few years later, another important status symbol-car-car identification has also become a hot topic. The general method of license plate recognition: computer image processing technology analyzes the license plate, automatically extracts the license plate information and determines the license plate number. Correlation coefficient of edge histogram and template matching hybrid algorithm based on vehicle identification. The recognition rate of off-line algorithm has reached a high level. At present, the theory of license plate logo pattern is relatively mature and is developing towards integration and intelligence.
In intelligent traffic management system, vehicle identification is equivalent to the "state" of v c++ base class, and other sub-modules inherit and develop on the basis of vehicle identification in intelligent traffic management system. Therefore, we think that automobile identification needs higher integration, and it is better to embed it into other systems and highly integrated modules, such as microcontroller and CPLD. At present, automobile identification is mostly done by computer.
In addition, due to the automatic identification and location of the base class, the use of "only cars can be locked" and "can quickly determine which cars will have certain requirements at this stage. Vehicle recognition, but only rely on simple license plate recognition. The theme of the market is a single license plate or vehicle identification system, and the combination of these two systems is very rare. It is obviously difficult for these single systems to truly identify the identity of locked cars.
Combined with the requirements of intelligent traffic management system, the present situation and two development trends of vehicle identification, the research group chose the embedded automatic vehicle identification system as an identity innovation experimental project. It is planned to complete the embedded processing of vehicle identification in the intelligent traffic management system and transfer the digital information to other modules, but the embedded computer will be used to process the vehicle number identification, which will greatly improve the integration of the intelligent traffic management system and reduce the cost. Different from a single recognition system, vehicle recognition system is designed for license plate recognition, and vehicle recognition is not combined, but supplemented by vehicle color recognition. At the same time, confirm and output the method of locking the car, and strive to be foolproof. This greatly facilitates the use of the system in various fields.
In the field of public security traffic management, it can be applied to embedded automatic identification systems, traffic control systems, tachometers used in embedded products, and other traffic facilities that measure overload, and can complete a series of management; The proces system connected to that terminal compute transmits processed digital information instead of image information, which greatly saves the proces time and storage space of the terminal computer, improves the response speed and processing efficiency, and effectively solves the manpower shortage problem in the field of traffic control.
In the vehicle management of the park, the automatic identification system embeds the identity of the outbound vehicle, so it can check whether the vehicle registered with the park owner is connected to the resource pool. Install an automatic license plate recognition system at the entrance of the park to automatically identify vehicles entering and leaving, and then check the data in the database with the license plate data in the database to determine whether it belongs to the parking lot, and then deal with it, which will greatly improve the car safety factor of the park. The cost of using this system is much lower than that of computer processing system.
Parking lot management and embedded automatic license plate recognition system can be completed in the process of intelligent management. The system is installed at the entrance of the parking lot and automatically identifies the vehicles in the parking lot. The processed data will be input into the database through the computer terminal to determine whether it is combined by the computer terminal to purchase (or rent) the motor vehicle parking space for corresponding processing.
In a word, we have reason to believe that the embedded automatic license plate recognition system we plan to complete can play a decisive role in the future intelligent traffic management system, which is worth studying and discussing.
BR/>;
Appendix II: Assumption of Engineering Scheme
Vehicle recognition system includes license plate recognition, vehicle color and vehicle main body recognition. The system will use embedded system to complete the identification. Because of this part of our content, this idea is not very mature.
For sub-license plate recognition and body color, logo pattern recognition is embedded in our works and programs.
: License plate recognition
1, overall structure
The automatic license plate recognition system is mainly divided into three modules: (1) trigger: the entrance speed measuring system of front-end equipment data. (2) Image processing: It is divided into four parts: image acquisition, license plate location, character segmentation and character recognition. (3) The wireless transmission system sends the processed data to back-end application systems, such as traffic violation management system, parking system and security system.
2. Algorithm part
① Before ending the CCD camera:
Original image acquisition
The image quality obtained by CCD camera and auxiliary lighting equipment will directly affect the effect of back-end processing and recognition. To get a clearer image, you need to consider many factors that affect the image quality, including: choosing a camera and an image acquisition card, calibrating the speed with the camera position, the access unit of the motorcade, the distance between the weather and light, the effect light, and the exposure of the camera.
Decide whether the vehicle enters the observation area.
The gray image difference method is used to judge whether the monitoring target area enters the first video image, and then the change times of the gray value of the corresponding pixel in the two images are compared, if there is any change.
The image difference can only be measured by monitoring the scene, and it remains to be seen whether it is a transport vehicle. After the noise is finished, the image is poor, pedestrians and bicycles account for more than cars, and the scale filter is designed to filter out smaller objects and noise.
② license plate location and preprocessing
Left license plate location algorithm. The basis of license plate location, but it also needs the basic preprocessing of license plate number.
Tilt straightening and boundary removal of rivets.
Me, license plate character tilt correction
In some license plates, it is difficult to directly segment license plate characters, which is invalid and needs to be corrected. First, we calculate the speed of license plate tilt and rotate the license plate tilt correction.
Two. Removal of license plate holder and rivet
Prior knowledge: For a standard license plate, the character spacing is 12 mm, the spacing between 2 and 3 characters is 34 mm, and the spacing between 2 and 3 characters with a width of 10 mm is12 mm. Generally, there are four rivets on the inner side of the license plate boundary line, and the first two words or the first six words are adhered to different degrees. If the rivets are not removed, the characters in 2 and 6 will be difficult to recognize. tail
The license plate image is binarized, and the image has only black and white binary files. White pixel (gray value 255), black pixel (gray value 0)0. Here, the license plate image is scanned line by line in black and white mode. When the width of the white pixel in the license plate image of the scanning line is greater than the threshold (the first qualified line), the license plate characters at the edge are removed, and all lines above or below this line are removed.
③ Segmentation of license plate characters
License plate algorithm displayed in the picture
Character segmentation.
Our limited knowledge
These algorithms are not described in detail.
④ Character recognition method .../> Character; Role; letter
Identify cars
brand recognition
Core part.
vehicle
Permission characters are known.
Include six non-grouping algorithms.
be included in
That's right.
We are more interested in the character recognition algorithm based on neural network. Next, we introduce two relatively simple and universal algorithms and a character recognition algorithm based on neural network.
First, template matching license plate character recognition
China license plate character templates are divided into Chinese characters, English letters and digital templates, and statistical methods are constructed and stored in the database. Use template matching character template and standardized license plate character matching to recognize characters.
Second, license plate character recognition and function matching
License plate recognition method has many personality characteristics, which can be roughly divided into structural characteristics, pixel distribution characteristics and other characteristics.
Here, we intend to break through the neural network method, because the nonlinear description of the large-scale parallel distributed processing ability, high robustness and self-learning association function of artificial neural network technology is suitable for the specific steps as follows:
In addition, we will try to combine various algorithms to avoid the simulation and online control of nonlinear time-varying large-scale systems. Weaknesses, such as the combination of genetic algorithm and artificial neural network, parallel calculation of genetic algorithm, can quickly use global search, and neural network can overcome the inherent shortcomings of slow speed and easy to fall into local dryness.
We are still a professional basic course in our sophomore year, and it is not enough to know the latest image processing algorithm. We will choose the best scheme and put forward improvement suggestions in combination with our system functions in actual operation. /a & gt;
Part two: the color and logo of the car.
① body color recognition
The size of color features depends on the tiny and powerful advantages of the image itself, direction and angle, and it has a very important application in content-based image retrieval technology and intelligent transportation system, as well as a large number of I-system industries (such as papermaking, textile, printing, etc.). For a long time, due to various reasons, a large number of color space models have been proposed, which can be mainly divided into three categories: the first category is based on human visual system (H VS), including RGB, H SI color space and M UNSELL color space; The second category is application-based color space, including YUV and YIQ used in TV system and photography industry, YCC and CMY(K) used in Kodak printing system; The third category is CIE color space (CIE XYZ, CIE Lab, CIE LUV, etc. The advantages and disadvantages of these color spaces and their important roles in their respective fields.
RGB color space, RGB intends to use our system. Color space is widely used in computer-related fields, such as common CRT displays, color values in RGB color space, and color values per? What is the combination of the values of R, G and B channels? Traditionally, the corresponding channel values pass through the photoreceptors in the image acquisition card or CCD sensor, and other similar devices. Where are the channel values? Incident light and its corresponding photosensitive function value and expression: R =
G =
B =
Where the frequency spectra of S(A), G(A) and R(A) and B(A) are sensors with sensitivity functions of R, G and B. As can be seen from the above formula, the color space is calculated in a computer, so it depends on the equipment and is associated with the photosensitive function of a specific capture device. However, because it is an easily available RGB value, it can usually be used to represent other color spaces, so what is the converted RGB value? The RGB color space standard color difference of other color space values is defined as:
)
People who have subjective feelings about different colors can better express:
The body color recognition system we intend to design mainly includes the following four steps.
1。 Selection of body color recognition area
In order to accurately identify the domain name? Selection and determination of body color. Facing the fan part of the car near the exhaust in the previous experiment.
2。 Color histogram calculation
Count the number of color occurrences in the selected area. In practical application, due to other component values? The color space model can be a simple calculation of RGB values, and the calculation of color histogram is only aimed at the RGB color space model.
3。 Color difference calculation
According to the color difference calculation formula of each color space model, the color template of color difference is calculated.
4. Color recognition
According to the results of the color space model of the sample color and the standard color, the aberration calculated in the previous step is selected as the lowest value of the recognition result in the corresponding components.
2. Don't identify the car.
Undeniably, real-time recognition of automatic license plate and vehicle subject is very important for accurate recognition of system and vehicle type. The proposed license plate location algorithm can be divided into two categories: license plate location algorithm based on black-and-white image and license plate location algorithm based on color image. Black and white images can be divided into many types, such as license plate location algorithm based on adaptive energy filtering, license plate location algorithm based on dyadic wavelet transform and morphological processing projection, and license plate location algorithm based on genetic algorithm.
License plate location algorithms have their own advantages and disadvantages, but to some extent, they are the benchmark for vehicle logo location.
The location and recognition of vehicle logo is a relatively new field at home and abroad. The inherent particularity of similar car logos: the target, size, lighting effect and background are not uniform, and the shapes and sizes of different car logos are inconsistent, which makes it difficult to accurately locate and identify.
License plate positioning is divided into vehicle identification, and its main steps are as follows:
(1) According to the texture features of the license plate, the license plate region can be quickly obtained on the basis of multi-resolution analysis;
(2) Pre-positioning: OTSU binarization image binarization algorithm, according to the front area? Higher energy is more and more concentrated, and then binary projection combined with license plate position information is used to quickly locate the front;
(3) Axis positioning: in the front area, position the front axle according to the axial symmetry;
(4) Rough positioning of automobile standard: based on the prior knowledge of automobile logo license plate before positioning, the empirical search rectangle of automobile logo is obtained;
(5) Step 1 (4) Based on the positioning accuracy of the vehicle logo, the vehicle body is accurately positioned by using the vehicle standard with texture features. The party consists of two steps: the car logo area has the characteristics of high energy and relative concentration in the vertical direction, the important part of the car identification system, the car license plate recognition, and the morphological filtering and adaptive car time positioning of the positioning recognition of two companies; Improved template matching algorithm for accurate positioning of vehicle logo. Key technologies. The picture shows the logo pattern of the car.
In other systems, a typical target recognition system includes a structural diagram of the recognition process during online and offline training. During the training process, the images collected manually by the automobile standard samples are normalized and standardized on a large scale, for example, before processing, in order to obtain the standard template library and template extraction of automobile logo. The template of the standard template library of automobile logo can not only locate automobile logo, but also be used for feature extraction. In the positioning process, in addition to the images to be imported, vehicles with license plate location information should also be input. Because all kinds of automobile standards do not have a stable texture feature, which is the same as the size and shape, it is very difficult to match the features or locate the logo directly with the template. Therefore, it is necessary to make full use of prior information, based on image processing technology and template matching, license plate positioning, symmetrical rough positioning and accurate positioning. The problem of vehicle identification after vehicle logo positioning is transformed into a 2D shape recognition problem, which can be realized by template matching. However, the actual image acquisition is often affected by light, noise and local occlusion, which forms similar problems. Traditional template matching methods are difficult to achieve satisfactory recognition results, so an appropriate feature extraction and recognition method is usually needed to assist vehicle identification and improve the recognition rate.
Part iii: embedded BR/> Based on a long history and universality, an embedded system should be defined as "a special computer system embedded in an object system." "Embedded", "private" and "computer system" are three basic elements of embedded system. The target system is embedded in the embedded system of the host system. core
Embedded system is an embedded microprocessor, which has four advantages:
(1) has a strong ability to support real-time multi-tasking, and can perform multi-tasking to minimize the interrupt response time, thus making the internal code and execution time of the real-time operating system;
(2) Powerful memory protection function.
(3) Extensible processor architecture, which can be rapidly expanded to meet the application of high-performance embedded microprocessors;
(4) The power consumption of embedded microprocessor is very low, especially in embedded systems that rely on battery power supply, especially in portable wireless and mobile computing and communication equipment, which is undoubtedly attractive in the era of increasingly scarce and expensive energy.
In addition, the embedded real-time operating system improves the reliability of the system. These are all worth making an embedded license plate recognition system.
Usually, the algorithm of license plate and vehicle identification considers a lot of calculations, but at the same time, it must meet the real-time requirements. Therefore, we will adopt 32-bit ARM embedded microprocessor as the core unit, CPLD as the timing control unit, embedded image acquisition and processing system based on ARM9sS3C24 1C, and embedded Linux operating system, and make full use of the characteristics of ARM's equipment, capabilities and low power consumption to realize parallel data bus /USB data interface image access, fast image processing, compressed local storage of image information and small-size data transmission based on IP. The system allows the whole system to simplify the circuit and reduce the resource intensity. compose
The system designs the U SB image acquisition subsystem, ARM processing subsystem and network data transmission subsystem of the whole system. The real-time video data collected by the camera is transmitted to the ARM processing board through USB, and the ARM processing board is embedded with Linux operating system. Fast imaging algorithm processes image sequences and takes appropriate measures. According to the processing results, the network transmission subsystem can process the data and upload it to the monitoring center for further follow-up processing. The system structure is shown in the figure.
The image processing subsystem of ARM adopts S3C 24 10 processor, and USB image access meeting the required image processing speed can ensure the image transmission speed, and expand 64M SD RAM and 64M flash memory. Large-capacity RAM can store multiple images, which is convenient for image analysis and processing and data network management of wireless network interface.
Of course, the above is only our preliminary idea, which was demonstrated and optimized in large-scale experiments!
Appendix III: Timetable
1。 /kloc-buy some basic experimental supplies in about 0/5 days.
2。 Make time to learn the necessary knowledge.
3。 Complete the program and solve the software problem in about 7 months.
In terms of hardware, it takes about a year to complete the prototype produced by the company.
5。
6。