The traffic sign recognition system of the car safety system, whose English translation is: Traffic Sign Recognition, or TSR for short, uses the front camera combined with the mode to recognize common traffic signs "speed limit, stop, U-turn" wait). This feature alerts the driver to traffic signs ahead so they can obey them. The TSR function improves safety by reducing the likelihood that drivers will disobey traffic laws such as stop signs and avoid illegal left turns or other unintentional traffic violations. These systems require flexible software platforms to enhance detection algorithms and adjust to traffic signs in different areas.
Traffic sign recognition, also known as TSR (Traffic Sign Recognition), refers to the ability to collect and identify road traffic sign information that appears while the vehicle is driving. , give prompt instructions or warnings to the driver, or directly control the vehicle to operate to ensure smooth traffic and prevent accidents. In vehicles equipped with safety-assisted driving systems, if the vehicle can provide an efficient TSR system, it can provide drivers with reliable road traffic sign information in a timely manner, effectively improving driving safety and comfort.
The following will introduce a typical road traffic sign recognition method.
TSR is based on the characteristics of objects recognized by the human visual system. Its recognition principle is to use the rich color information and fixed shape information of road signs for feature recognition. Specifically, the recognition process can be divided into two steps: "separation" and "recognition". Separation refers to finding candidate targets in the acquired images and performing corresponding preprocessing, followed by traffic sign recognition, including feature extraction and classification, and finally further determining the authenticity of the target.
Traffic sign separation actually requires quickly obtaining interesting information that may be traffic signs from complex scene images. area. Then the pattern recognition method is used to further identify the area of interest and locate its specific location. Since traffic signs function as indicators, reminders and warnings, they are designed to be eye-catching, bright in color, concise in graphics, and clear in meaning. Therefore regions of interest are usually mapped using their color and shape.
Currently, the color spaces commonly used in traffic sign recognition include RGB, HIS, and CIE. RGB, as the three primary colors commonly used in image processing, is the basis for constructing various other colors. Others Color representation can be obtained by RGB conversion.
We know that for traffic signs, most colors are relatively single and fixed. For example, red signs generally indicate prohibitions, blue signs generally indicate instructions, and yellow signs generally indicate instructions. Generally speaking, it represents a warning type. Here, the three primary colors of RGB, red, yellow, and blue, are used to identify and match them.
Because the color space also contains more information such as saturation, in order to better distinguish color and brightness information, researchers use more color models suitable for human visual characteristics HSI models to process traffic For logo recognition, H represents different colors, S represents color depth, and I represents the degree of lightness and darkness. The biggest feature of HSI is that there is minimal correlation between H, S, and I. Each color image in the HSI space corresponds to a relative Consistent shade H.
When the area of interest of the traffic sign information is segmented in the test driving environment, a certain algorithm needs to be used Discriminate it in order to determine which specific traffic sign it belongs to. General discrimination methods include template matching methods, cluster analysis-based methods, shape analysis-based methods, neural network analysis methods, and support vector machine-based methods.
(1) Based on template matching method
(2) Based on cluster analysis method
(3) Method based on neural network
(4) Support vector machine method
Support vector machine is a typical feedforward neural network method used to solve pattern classification and nonlinear problems. The main idea is to establish an optimal decision-making hyperplane to maximize the distance between the two types of samples closest to the plane on both sides of the plane, providing better generalization capabilities for classification. For nonlinear separable pattern classification problems, the responsible pattern classification problem needs to be nonlinearly projected into a high-dimensional feature space. Therefore, as long as the transformation is nonlinear and the dimensionality of the feature space is high enough, the original pattern space can become a new A high-dimensional feature space in which patterns become linearly separable with high probability. The transformation process requires generating a kernel function for convolution. The corresponding typical kernel function is expressed as follows:
Gaussian function: picture; used for radial set function classifier;
Inner product function: picture; used for high-order polynomial set classifier;
Sigmoid kernel function: picture; used to implement a single hidden layer perceptron neural network.
Because the complexity of road traffic conditions may cause traffic signs to be stained, their colors and shapes to change, and the appearance of trees and buildings Occlusion may cause it to be unable to be recognized in time. At the same time, during high-speed driving, factors such as vehicle jitter may cause errors in the image frame matching process, making it impossible to stably recognize the corresponding traffic signs. Therefore, traffic sign recognition has not yet been widely used in the field of driving assistance. The more mature application solutions include the following:
The automatic speed limit based on the speed limit sign mainly uses the speed limit value displayed by the recognized speed limit sign, and the vehicle predicts it in advance. Here we set several different speed values for comparison.
VReal represents the current actual cruising speed of the vehicle, Vtarget represents the target cruising speed of the vehicle, Vlim represents the speed limit value information, and Vfront represents the recognized speed of the vehicle ahead.
Based on the sensitive information of your own speed, the following speed limit strategies are implemented to varying degrees:
1) Cruise control of this vehicle
When it is detected that the vehicle VReal>Vlim and Vtarget
When it detects that the vehicle VReal
2) The vehicle follows the vehicle in front
When detecting VReal>VFront>Vlim of the vehicle, the system will ensure that the vehicle does not collide with the vehicle in front. Automatic deceleration control;
When detecting that the vehicle VReal
3) Control logic after passing the speed limit sign
After the above vehicle automatically limits the speed, when the vehicle passes the speed limit sign, it will recognize the new speed limit sign at the same time. It is necessary to re-control the speed when speeding. If the speed limit value of the new speed limit plate is smaller than the current value, further speed restriction will be carried out according to the logic in 1) 2). If the speed limit value of the new speed limit plate is greater than the current value, , then the acceleration needs to be redistributed based on the current updated actual speed of the vehicle, the speed of the vehicle in front, and the target cruise speed of the vehicle to ensure that speed limits and collisions are prioritized and properly controlled.
For vehicles driving on highways, driving vehicles need to consider There are currently two feasible solutions for the problem of early lane changes in different scenarios:
First, when a lane merging sign information is detected ahead at a certain distance, the vehicle can be changed in advance through voice or instrument images. Prompt the driver to control the lane change of the vehicle and change the vehicle to the target lane;
Secondly, if the system receives lane level information related to high-precision map before a certain distance, it can directly control the vehicle When changing lanes to the target lane, it is necessary to detect whether the target lane line is a dotted line, whether the target lane is safe to change, etc.
The driving assistance system based on traffic light sign recognition requires the system to control vehicle cruising and lane changing in advance based on the recognized traffic lights. .
There are mainly the following control scene strategies:
1) When the green light is recognized
If the vehicle is following the vehicle in front , the vehicle in front is driving at a lower speed. On the premise of ensuring collision safety, the vehicle continues to follow the vehicle in front and monitors the changes in the lights in real time. Once the light turns yellow, it will immediately stop following the strategy and maintain a certain deceleration to Brake;
2) Recognize the yellow light
If the vehicle recognizes the yellow light, it is required regardless of whether the vehicle is following the vehicle or not. Control the vehicle's deceleration and stop. During the deceleration process, you can decelerate for comfort, and switch from engine reverse drag to brake cut-in;
3) Recognize the red light
If the red light has been recognized, according to the stopping state of the vehicle in front, and on the premise of ensuring collision avoidance, the vehicle will be controlled to decelerate to a stop and maintain a distance of more than 1m from the vehicle in front of the vehicle in the braking state;
(Among them, the application of the first function is easier to understand; the application of the second function is more significant for development, and it involves the automatic lane change logic from L2 to L3 level; the application of the third function , seems smarter - braking in advance, similar to the logic of V2X.)
Combining the development of the technology itself and China's road traffic environment and the specific needs of consumers, we can summarize the development trends of advanced driving assistance system technology in the Chinese market:
(1) From the perspective of technological development, as consumers will only pay more and more attention to automobile safety, advanced driving assistance systems will maintain a continuous development trend for a long time to come. At the same time, advanced driving assistance systems are changing from the independent development of a single technology to the development of integrated active safety systems. Multiple technologies can share platforms such as sensors and control systems. Once the vehicle is equipped with basic ESP, ACC and other technologies, it can be easily And adding other safe driving assistance technologies at a lower cost will further promote the application of advanced driving assistance system technologies in automobiles.
(2) Some relatively low-end and highly practical advanced driving assistance system technologies, such as tire pressure monitoring systems, ESP electronic stability systems, etc., have been fully recognized by the market. Under the strong pressure Driven by demand, its penetration rate in the low-end market will steadily increase.
(3) Chinese consumers have shown obvious attention and demand for hedging assistance and vision improvement technologies, which will surely become the main growth point in this field in the next stage.
(4) Some technologies with higher road requirements, such as lane change assist, lane departure warning, ACC, etc., as well as technologies that are inconsistent with the driving habits of Chinese consumers, such as lane keeping systems, driving Staff fatigue detection, alcohol ban lockout system, etc. may face slow development for a long period of time.
Today, current technology cannot determine all traffic signs, nor can it operate under all conditions. There are several conditions that limit the performance of the TSR system, including the following:
While TSR and similar vehicle sensing technologies are helpful in moving toward fully autonomous driving, we are not there yet. Even TSR is just a driving assistance system. Drivers cannot rely entirely on any ADAS system to drive for them.
In general, the basic functions of TSR are relatively mature, but there is still some way to go for advanced functions and the simplification of the ecological chain.
The above is the detailed content of An article talking about the traffic sign recognition system in autonomous driving. For more information, please follow other related articles on the PHP Chinese website!
![[Web front-end] Node.js quick start](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
