I found it, haha. Personal information will not be uploaded, only key ones.
Embedded Automobile Identity Automatic Identification System
1. Project Introduction
(Research objectives, research background and current situation, working principle and scheme assumptions, plan schedule etc.)
See appendix.
II. Project self-evaluation
1. Advancement:
In the post-PC era with the rapid development of digital information technology and network technology, with the rapid development of embedded processing With the continuous improvement of processor performance, high-performance processors can already meet complex algorithm applications and other complex function applications. Embedded will inevitably enter various fields. On the other hand, with the rapid development of our country's economy and the holding of the Beijing Olympics, "intelligent transportation" will undoubtedly become a hot topic. Due to the particularity of the transportation industry, it has strict requirements on the technical parameters and usage conditions of ITS equipment, and embedded can just meet this requirement. Therefore, the wide-scale application of embedded intelligent transportation equipment is an inevitable trend. The embedded automatic car identity recognition system is an important part of the intelligent traffic management system. It is the perfect combination of embedded technology and car identity recognition technology. It covers three major categories: embedded license plate recognition, embedded car logo recognition and car color recognition. The main function strives to lock the car target at once.
It has the following advantages:
1. Highly independent: It uses embedded technology and is connected to the application system only through the communication interface, with high independence.
2. Complete functions: It can simultaneously identify car license plates, logos and colors, lock the target at once, and has powerful functions that existing systems do not have.
3. Strong plasticity: the front end can be combined with upstream products such as signal triggering devices, and the end has built-in wireless networks and multiple serial interfaces for integration with downstream products. System functions and scope of use have been greatly expanded.
4. Easy to maintain: Repair and maintenance only involve this system and do not affect other modules. The maintenance cost is much lower than similar products.
5. Portable and flexible: The device is highly integrated, compact and flexible, and easy to use.
2. Operability and realizability:
At present, technologies such as license plate recognition and vehicle logo recognition are becoming increasingly mature and perfect, and relevant information is easier to obtain. Existing embedded technology is also relatively mature. Therefore, in terms of technical difficulty, this topic is easier to implement than other cutting-edge sciences. The equipment and materials involved in the topic selection are also relatively easy to obtain and the cost is moderate.
3. Innovation:
Existing license plate recognition devices generally use computers to process data, and some even require the cooperation of several computers, which takes up a lot of space and resources. Even if there are occasional embedded systems, their functions are limited to license plate recognition or vehicle logo recognition. This system creatively combines embedded with license plate recognition, car logo recognition and car color recognition, solving the current problems of bloated equipment system, difficult integration, poor stability, difficult maintenance and single function in one go.
4. Possible problems:
Currently, the main problems are embedded integration and wireless transmission distance. The ideal situation we envisage is to develop a portable car identification system that can transmit data wirelessly to address the shortcomings of the current use of complete computers to process data and poor equipment flexibility. But due to the limitations of our time, energy and funds, "portability" is currently the biggest problem. In addition, the impact of vehicle speed and depth of field on image recognition is also a difficult problem we may face.
3. Expected results
(The specific form of results, such as: applying for patents, publishing papers, producing scientific and technological objects (including software programs), etc., can have multiple results forms at the same time )
We expect our experimental results to have the following aspects.
First of all, we plan to produce physical technology, that is, to actually complete the embedded system and produce tangible results.
Second, from our analysis of the current market situation, the market prospects of the embedded car identification system are very optimistic, so we can apply for a patent for our product and put it into the market for production.
Thirdly, since there is no suitable algorithm for the combined recognition of car color, license plate, and car logo, we inevitably have to complete the algorithm design in the process of completing this system, and this part of the results can be achieved through Presented in the form of a published paper.
Because we plan to complete a system, we need to complete both the hardware and software parts of the system at the same time. From a large perspective, the results of software and algorithms can be published through papers, while the results of hardware can be reflected by putting into production and applying for patents. Undoubtedly, our results will be in more forms than those that only do the software part or only the hardware part. This is also one of our great advantages.
Experimental environment requirements
Fund budget content, purpose, budget amount, estimated execution time
Acquisition of front-end images of the CCD camera part, purchase of camera or camera 3000 07.12~ 08.2 months
Auxiliary light source to supplement light for special environments 1500 07.12~ 08.2 months
Image acquisition card analog signal digitization 2500 07.12~ 08.2 months
Embedded system hardware facilities Image processing 4000 08.3~ 08.10 months
Storage of hard disk video recorder video information 2500 08.10~ 08.12 months
Display device outputs image recognition results 1500 08.12~ 09.2 months
Transmission of information from wireless transceivers or wired transmission devices 2500 09. 2~ 09.3
Mechanical processing of mechanical parts and assembly into prototypes 2000 Final stage
Total: 19,500 yuan
< p>College Approval OpinionsExpert Committee Approval Opinions
School Approval Opinions
Appendix 1: Current Situation, Background and Significance of Topic Selection
Since the birth of the world's first automobile in 1885, automobiles have had a tremendous impact on our daily work and life. Over the past 100 years, automobiles have gradually been accepted by the public due to their low prices and convenient operation, and have entered thousands of households. In our country, many people join the car ownership class every year. Naturally, what follows is an increasingly fast and convenient lifestyle and a series of problems caused by it: more than 10,000 car thefts occur every year, and traffic accidents occur from time to time... There is no doubt that cars need to be regulated and managed. Nowadays, most of the automobile management work in our country is performed by people. It is not difficult to imagine that in the face of an increasingly large fleet of cars, manual operations are obviously insufficient. Therefore, "traffic intelligence" will become an inevitable trend in future traffic management.
How can we achieve intelligent transportation without the identification of "car identity". As early as the early 1990s, car identification has attracted widespread attention around the world, and people began to study related issues related to car ID cards-automatic recognition of car license plates. A few years later, another important status symbol of cars - car logo recognition has also become a hot topic. The general approach to license plate recognition is to use computer image processing technology to analyze the license plate and then automatically extract the license plate information to determine the license plate number. Vehicle logo recognition is based on an algorithm that mixes edge histograms and template matching correlation coefficients. At present, the theory of license plate and vehicle logo identification has matured, the offline algorithm recognition rate has reached a high level, and it is developing in the direction of integration and intelligence.
In the intelligent traffic management system, car identification is equivalent to the "base class" status in vc++, that is, other sub-modules in the intelligent traffic management system need to inherit on the basis of car identification. and development. Therefore, we believe that the high level of integration required for automotive identity recognition can best be accomplished by highly integrated modules that can be embedded in other systems, such as microcontrollers and CPLDs. At this stage, most of the car identification is done by computers.
In addition, due to the "base class" positioning of car identification, there are certain requirements for "whether it can uniquely lock the car" and "whether it can quickly determine which car it is" when using it. . At this stage, car identification only relies on simple recognition of license plates. Most of the systems on the market are separate recognition systems for license plates or vehicle logos, but systems that combine the two are very rare. It is obviously difficult for these single systems to achieve the purpose of truly identifying and locking the car's identity.
Based on the requirements of intelligent traffic management systems, the current status of automobile identification and the development trends of both, our team chose the embedded automatic identification system for automobiles as the topic of our innovative experimental plan. . We plan to complete the car identification in an embedded manner and then transfer the processed digital information to other modules of the intelligent traffic management system. Using embedded computers instead of computers to process vehicle identification will greatly improve the integration of intelligent traffic management systems and reduce costs. Different from a single identification system, the car identification system we designed combines license plate recognition with vehicle logo recognition, supplemented by car color recognition. Simultaneous identification and simultaneous output are used to judge and lock the car from multiple aspects, striving to be foolproof. This greatly facilitates the use of the system in various fields.
In the field of public security traffic control, the embedded automatic car identity recognition system can be applied in traffic control systems. By embedding this product into other transportation facilities used to measure speed and overload, a series of management tasks can be completed; connected to the terminal computer processing system, the processed digital information is transmitted instead of picture information, which greatly saves money. The processing time and memory space of the terminal computer improves the response speed and processing efficiency, effectively solving the current situation of insufficient manpower in the traffic management field.
In terms of vehicle management in the park, this embedded automatic car identification system will have a port so that it can be connected to the car information database registered by the park owners when they check in. At the gate of the park, install our automatic license plate recognition system to automatically identify vehicles entering and exiting, then transfer the data to the database and determine whether it is a vehicle in the park based on the license plate data in the database, and then handle it according to the situation. This will greatly increase the safety factor of the park's cars, and the cost of using the system is much lower than the cost of using a computer-processed system.
As for parking lot management, our embedded automatic license plate recognition system can complete the intelligent management process. The system is installed at the entrance and exit of the parking lot to automatically identify vehicles entering and exiting the parking lot. The processed data will be transmitted to the terminal computer. The terminal computer will combine the incoming information and the database to determine whether the vehicle has been purchased. (or rent) parking spaces will be handled accordingly.
To sum up, 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 transportation management system and is worthy of research and exploration.
Appendix 2: Working principle and scheme idea
This car identification system includes license plate recognition, car color and car logo recognition. This system will use an embedded system to complete these three parts. Identify. Since we are new to this part of the content, our ideas are not very mature.
The following will introduce our working principles and solutions in three parts: license plate recognition, car color, car logo recognition and embedded.
Part One: License Plate Recognition
1. Overall Structure
The automatic license plate recognition system is mainly divided into three modules: (1) Trigger: that is, the front-end device Data entrance, such as speed measurement system, etc. (2) Image processing part: It is divided into four parts: image collection, license plate positioning, character segmentation and character recognition. (3) The wireless transmission system transmits the processed data to back-end application systems, such as traffic violation management systems, parking systems, security inspection systems, etc.
2. Algorithm part
①Front-end CCD camera:
Original image acquisition
It is composed of CCD camera and auxiliary lighting device. The quality of the acquired image directly affects the effect of back-end processing and recognition. To obtain a clearer image, many factors affecting image quality need to be considered, mainly including: the selection of the camera and image card, the position calibration of the camera, and the position of the car. The influence of vehicle speed, distance between the car fleet entering and exiting the unit, weather, light and other conditions on the exposure of the image captured by the camera.
Determine whether a vehicle has entered the observation area
Use the image difference method to determine whether a target has entered the observation area, that is, first grayscale the video image, and then compare the correspondence between the two images Check the grayscale value of the pixel to see if there is any change and how much the change is.
Image difference can only determine whether there is an object passing through the monitoring area, but it is not yet known whether it is a traffic vehicle.
In view of the fact that the noise generated by image difference, pedestrians, and bicycles occupy a much smaller area than cars, a scale filter is designed to filter out smaller objects and noise.
②License plate positioning and preprocessing
The picture on the left shows the main algorithm for license plate positioning. After completing the basic license plate positioning, some basic preprocessing of the license plate is required. Includes tilt correction
with rivet and frame removal.
I. Slant correction of license plate characters
The difficulty in segmenting license plate characters is that some license plates are tilted, and direct segmentation is not effective and requires correction. First, find the slope of the license plate, and perform rotation correction on the license plate based on this slope.
II. Removal of license plate frame and rivets
Prior knowledge: For standard license plates, the spacing between characters is 12mm, and the spacing between the second and third characters is 34mm. Among them, the middle The small dot is 10mm wide, and the spacing between the small dot and the second and third characters is 12mm respectively. There are usually four rivets on the inside of the license plate border line. They are adhered to the 2nd or 6th character to varying degrees. If the rivets are not removed, it will cause difficulty in identifying the 2nd and 6th characters.
After binarizing the license plate image, the image only has black and white values. White pixels (grayscale value 255) take 1, and black pixels (grayscale value 0) take 0. The black text on white background mode is used here. The license plate image is scanned line by line from the inside to the outside. When a certain line of the license plate image is scanned, and the width of the white pixel is greater than a certain threshold (the first line that meets the conditions), it is considered to be at the edge of the license plate characters. , excise all rows above or below this row.
③License plate character segmentation
The picture on the right shows the main algorithm of license plate
character segmentation
.
Here, due to
our limited knowledge
we will not introduce these algorithms
in detail.
④Character recognition method
Character
recognition is the core part of car plate recognition
.
Common car license plate character recognition
Algorithms include
six types.
We list them
in the picture on the right.
Among them, we are more interested in the character recognition algorithm based on neural networks. Below, we introduce in detail two relatively simple and common algorithms and character recognition algorithms based on neural networks.
I. Template matching license plate character recognition
The character templates of Chinese license plates are divided into Chinese characters, English letters and numerical templates, which are constructed by statistical methods and saved in the database. Template matching is to match character templates with standardized license plate characters to identify characters.
II. Feature matching license plate character recognition
In the method of license plate recognition, there are many character features that can be used, which can be roughly divided into structural features, pixel distribution features and other features.
Here, we plan to focus on breaking through the neural network method, because artificial neural network technology has the characteristics of non-linear description, large-scale parallel distributed processing capabilities, high robustness, self-learning and association, and is suitable for non-linear applications. Simulation and online control of linear time-varying large systems. The specific steps are shown in the figure below:
In addition, we will also try to combine various algorithms to maximize strengths and avoid weaknesses, such as combining genetic algorithms with artificial neural networks, which can not only use genetic algorithms to perform parallel calculations Moreover, the advantages of fast and global search can overcome the inherent shortcomings of neural network such as slow search speed and easy to fall into local drought and heat.
Since we are still studying professional basic courses in the second year of college and do not know enough about the latest algorithms for image processing, we will choose an optimal solution and combine it with our system characteristics during the actual operation. Provide suggestions for improvements.
Part 2: Car color and car logo recognition
①, Car body color recognition
Color features are dependent on the size, direction, viewing angle, etc. of the image itself It has the advantages of small size and high robustness, so it has extremely important applications in content-based image indexing technology and intelligent transportation systems as well as many industrial (such as papermaking, textile, printing, etc.) systems. For a long time, due to various reasons, people have proposed a large number of color space models, which can be mainly divided into three categories: The first category is the color space based on the human visual system (Human Vision System, H VS), which includes RGB, H SI, M unsell color space, etc.; the second category is a color space based on specific applications, which includes YUV and YIQ adopted in television systems, YCC in the photography industry such as Kodak, and CMY (K) color space in printing systems; the third category The class is CIE color space (including CIE XYZ, CIE Lab and CIE Luv, etc.). Each of these color spaces has advantages and disadvantages, and they play an important role in their respective fields.
We plan to use RGB color space to complete our system. RGB color space is widely used in computer-related fields, such as common CRT monitors. In the RGB color space, each color value is represented by a combination of R, G, and B channel values, and the corresponding channel value is obtained through a photoreceptor in a graphics acquisition card or a CCD sensor or other similar device. of. Among them, the value of each channel is represented by the sum of the incident light and the photosensitivity function value of the corresponding photoreceptor:
R=
G=
B= < /p>
Where, S (A) is the spectrum, R (A), G (A) and B (A) are the sensitivity functions of the R, G and B sensors respectively. As can be seen from the above equation, this color space is device-dependent and is related to the photosensitivity function of the specific capture device. However, because RGB values ??are easy to obtain and can be calculated and expressed in computers, they can usually be used to represent other color spaces, that is, to convert RGB values ??into other color space values. The standard color difference of RGB color space is defined as:
)
Since different colors have different effects on people’s subjective feelings, in order to better express the color difference, in this color recognition subsystem Use the empirical color difference formula:
The car body color recognition system we plan to design mainly divides the following four steps to complete the car body color recognition
1. Selection of recognition area
In order to accurately recognize the color of the car body, the selection of the recognition area is very important. This experiment selected the front part of the car face close to the exhaust fan
2. Color histogram calculation
For the selected area, calculate the color that appears most frequently. In practical applications, since the component values ??of other color space models can be represented by RGB values, for simplicity of calculation, the color histogram can be calculated only for the RGB color space model.
3. Color difference calculation
According to the color difference calculation formula of the corresponding color space model, calculate the color difference between it and the color template.
4. Color identification
After obtaining the corresponding color difference between the sample color and the standard color in each color space model, color identification can be carried out based on the results. That is, the minimum value among the color differences calculated in the previous step is selected as the recognition result.
②. Vehicle logo recognition part
There is no doubt that the automatic and real-time recognition of license plates and vehicle logos are two crucial parts in the accurate identification system of sports vehicle types. At present, many license plate positioning algorithms have been proposed, which can be mainly divided into two categories: license plate positioning algorithms based on black and white images and license plate positioning algorithms based on color images. License plate location algorithms based on black and white images can be divided into many categories, such as feature-based license plate location algorithms, license plate location algorithms based on adaptive energy filtering, license plate location algorithms based on a combination of wavelet transform and morphological processing, license plate location algorithms based on binary projection License plate positioning algorithm, and license plate positioning algorithm based on genetic algorithm, etc.
These license plate positioning algorithms each have their own advantages and disadvantages, but they can all be used as a reference for vehicle logo positioning to a certain extent.
Vehicle logo positioning and identification is a relatively new field both at home and abroad.
Due to the inherent particularities of the car logo itself: small target, large similarity, large influence by size and lighting, non-uniform background, and inconsistent shape and size of car logos of different car companies, etc., making its precise positioning and recognition a difficult point.
We divide vehicle logo recognition into the following main steps:
(l) License plate positioning: According to the texture characteristics of the license plate, the license plate area is quickly obtained based on multi-resolution analysis;< /p>
(2) Car front positioning: According to the characteristics of high and concentrated energy in the front area, the image is binarized through the OTSU binarization algorithm, and then binary projection is used, combined with the license plate position information, to quickly determine the front of the car. Positioning;
(3) Center axis positioning: In the front area of ??the car, position the center axis of the front according to axial symmetry;
(4) Rough positioning of the car logo: After locating the front of the car Basically, based on the prior knowledge of the vehicle logo and license plate, the empirical search rectangle of the vehicle logo is obtained;
(5) Accurate positioning of the vehicle logo: Based on step (4), use the texture features of the vehicle logo Carry out precise positioning of vehicle logos. It mainly includes two steps: first, according to the characteristics of high energy and relatively concentrated energy in the vertical direction of the vehicle logo area, use energy enhancement and adaptive morphological filtering to perform primary positioning of the vehicle logo; second, use an improved template matching algorithm to perform vehicle logo positioning Pinpoint positioning. The vehicle logo recognition system is an important part of the sports vehicle recognition system. Like license plate recognition, it also includes two key technologies: positioning and recognition.
The above picture is a schematic structural diagram of the vehicle logo recognition system. Like a typical target recognition system, it includes an offline training process and an online recognition process. During the training process, the manually collected vehicle logo samples are first subjected to preprocessing such as image normalization and scale normalization, and then the templates are extracted separately to obtain the vehicle logo standard template library. The templates in the vehicle logo standard template library are not only used for vehicle logo positioning, but also for feature extraction to obtain a vehicle logo feature model library for vehicle logo recognition. During the positioning process, in addition to inputting the car image, the location information of the license plate also needs to be input. This is because various types of vehicle logos do not have stable texture features and vary in size and shape. Therefore, it is very difficult to directly use feature matching or template matching to locate vehicle logos in complex backgrounds. Therefore, prior information such as license plate position and vehicle symmetry must be used for rough positioning, and on this basis, relevant image processing technology and template matching can be used for precise positioning. After the vehicle logo is positioned, the vehicle logo recognition problem is transformed into a 2D shape recognition problem, which can be achieved through template matching. However, in actual collected images, there are often problems such as illumination, noise, partial occlusion, and similar shapes, making it difficult for conventional template matching methods to achieve satisfactory recognition results. Therefore, a suitable feature extraction and recognition method is usually needed to assist vehicle logo recognition to improve the recognition rate of the system.
Part Three: Embedded
According to the requirements of history, essence, and universality, embedded systems should be defined as: "Special computer systems embedded in the object system." "Embeddedness", "specificity" and "computer system" are the three basic elements of embedded systems. The object system refers to the host system in which the embedded system is embedded.
The core of the embedded system is the embedded microprocessor, which has four advantages:
(1) It has strong support for real-time and multi-tasking and can complete multi-tasking. And it has a short interrupt response time, thereby reducing the execution time of the internal code and real-time operating system to a minimum;
(2) It has a powerful storage area protection function.
(3) The scalable processor structure can quickly expand high-performance embedded microprocessors that meet the application;
(4) Power consumption of embedded microprocessors Very low, especially for battery-powered embedded systems used in portable wireless and mobile computing and communication equipment. The power consumption can only be in the mW or even μ W level. This is especially true in an era where energy is becoming increasingly scarce and expensive. , is undoubtedly very tempting.
In addition, the embedded real-time operating system improves the reliability of the system. These are all worthy of us building an embedded license plate recognition system.
Considering that the usual license plate and vehicle logo recognition algorithms require a large amount of calculations and must meet real-time requirements.
Therefore, we plan to use a 32-bit ARM embedded microprocessor as the core unit, CPLD as the timing control unit, and an embedded image acquisition and processing system based on ARM 9 S3C 241 C. Based on the embedded Linux operating system, we fully Taking advantage of the small size, strong capabilities and low power consumption of ARM devices, it realizes parallel data bus/USB interface image access, rapid image processing, local compressed storage of image information and IP digital data transmission. This system can simplify the circuit and reduce the occupied resources of the entire system.
System design and composition
The entire system consists of a USB image acquisition subsystem, an ARM processing subsystem and a network data transmission subsystem. The camera collects on-site video data and transmits it to the ARM processing board through USB. ;The ARM processing board is embedded with the Linux operating system, using fast image algorithms to process the image sequence, and taking corresponding measures based on 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 below .
The ARM image processing subsystem plans to use the S3C 2410 processor, which can meet the image processing speed requirements; USB image access can ensure the image transmission speed; expand 64M SD RAM and 64M Flash, large-capacity RAM It can save multiple images to facilitate image analysis and processing; the wireless network interface realizes network management of data information.
Of course, the above are just our preliminary ideas. These ideas will be demonstrated and optimized in our future experiments!
Appendix 3: Plan progress and arrangement
Plan progress and arrangement:
1. Spend about 15 days buying some basic supplies needed for the experiment.
2. Use your spare time to learn the required knowledge.
3. It takes about seven months to complete programming and solve software problems.
4. It will take about one year to complete the hardware aspects and make a prototype.
5. The initial inspection takes about one month.
6. Debug the prototype for six months, find defects and correct them. Experiment and repeat until you reach a satisfactory level.
To sum up, we plan to win this project in about two years. Of course, the above is just a general plan and will be adjusted appropriately in the future according to the actual progress of the experiment.