Intelligent curve early warning system and method

Abstract

The invention provides a curve intelligent early warning system and method, and the system comprises a sensing layer which comprises a sensor and a camera, and is used for collecting the speed data, position data and vehicle image data of bidirectional vehicles in a curve region; the transmission layer comprises a network module and a display device, the network module is used for transmitting the data collected by the sensing layer to the application layer, and the display device is used for displaying curve early warning information; the application layer is used for decoding and classified storage of data; the platform layer communicates with the application layer, analyzes data decoded by the application layer in real time, identifies vehicle image data information, and sends a curve early warning signal to a display device when the data reaches a threshold value; and the facility layer is a cloud resource module, communicates with the application layer and is used for storing and cleaning the data.

Description

Technical Field

[0001] The present invention relates to the field of traffic safety, and in particular to an intelligent curve warning system and method. Background Art

[0002] At present, there are two main methods for the research on intelligent safety warning technology for internal roads in scenic spots:

[0003] One is a curve crossing warning system that relies on geomagnetic detection. The system has geomagnetic detection devices and warning devices at the entrances on both sides of the curve. The geomagnetic detection devices are used to detect whether a vehicle enters the curve. When a vehicle is detected entering the curve, the warning information is displayed on the warning display device on the other side.

[0004] Another method is a multi-curve road section safety warning system that relies on two communication subsystems. The system divides the multi-curve road section safety warning system into a multi-curve road section roadside safety warning system and a multi-curve road vehicle safety warning system. The multi-curve road section roadside safety warning system communicates data with the multi-curve road vehicle safety warning system, thereby achieving a warning effect on vehicles passing through the curved road section.

[0005] The existing curve warning technology has problems with inaccurate recognition of the vehicle's curve driving status and slow update of warning information, resulting in low curve warning efficiency. Summary of the invention

[0006] In order to make up for the deficiencies in the prior art, the present invention provides an intelligent curve warning system and method, which can effectively improve the efficiency of curve warning.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] Intelligent curve warning system, including:

[0009] The perception layer includes sensors and cameras, which are used to collect speed data, position data, and vehicle image data of bidirectional vehicles in the curve area;

[0010] The transmission layer includes a network module and a display device, wherein the network module is used to transmit the data collected by the perception layer to the application layer, and the display device is used to display the curve warning information;

[0011] Application layer, used for decoding and classified storage of data;

[0012] The platform layer communicates with the application layer, analyzes the data decoded by the application layer in real time, identifies the vehicle image data information, and sends a curve warning signal to the display device when the data reaches the threshold value;

[0013] The facility layer, which is a cloud resource module, communicates with the application layer and is used for data storage and cleaning.

[0014] Furthermore, the sensor includes a light sensor and a radar.

[0015] Furthermore, the display device includes an LED display screen and a warning light.

[0016] Furthermore, the curve warning information includes license plate number, vehicle type, vehicle distance, and vehicle speed.

[0017] The present invention also provides a curve intelligent warning method, comprising the following steps:

[0018] S1. After the radar in the sensing layer senses that the vehicle enters the curve area, it collects the vehicle speed and position data and triggers the camera to start;

[0019] S2. The camera captures the vehicle and obtains vehicle image data;

[0020] S3. The data collected by radar and camera are transmitted through the transmission layer and application layer to the platform layer, which performs data analysis, algorithm recognition, and data statistics on the data;

[0021] S4. After the platform layer analyzes and identifies the curve warning information, it sends it to the display device for display;

[0022] S5. The warning lights flash in different types according to the distance and speed of the vehicle to warn the vehicle of curves.

[0023] Furthermore, in step S3, the data acquired by the platform layer is cleansed by the facility layer.

[0024] Furthermore, the data cleaning adopts Hdop technology.

[0025] Furthermore, in step S3, the platform layer uses edge computing and multi-target recognition technology to capture information of vehicles traveling on curves and identify the license plate number, vehicle type, and driver's facial information.

[0026] Compared with the prior art, the present invention has the following beneficial technical effects:

[0027] The present invention can realize real-time and accurate perception and data analysis of the vehicle position, speed and driving status in the curve area, can provide early warning for two-way lanes, and combine with algorithm models to increase the diversity of early warning methods and improve the efficiency of curve early warning.

[0028] The curve warning system mainly involves the fields of traffic safety and driving assistance. Its benefits are mainly reflected in improving road safety, reducing traffic accidents, improving driving efficiency and enhancing traffic smoothness, rather than being directly related to reduced energy consumption, increased yield, increased purity, simplified processes, reduced costs or reduced pollution.

[0029] However, from an indirect perspective, it can produce some positive chain reactions, such as:

[0030] Improve traffic efficiency: With early warning, drivers can pass through curves more smoothly, reducing traffic congestion or accidents caused by improper operations, thereby improving road traffic efficiency overall.

[0031] Reduced energy consumption: If the curve warning system can help vehicles pass through curves at a more stable speed and avoid frequent braking and acceleration, this may save fuel and reduce energy consumption to a certain extent.

[0032] Reduce costs: Reducing traffic accidents means reducing the economic losses caused by them, including vehicle repair costs, medical expenses, and insurance costs, which is a cost saving for both individuals and society.

[0033] Environmental protection: Since unnecessary acceleration and braking are reduced, vehicle emissions may also be reduced, helping to reduce environmental pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the training model of LSTM long short memory built by the present invention. DETAILED DESCRIPTION

[0035] Many specific details are described in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from the description, and those skilled in the art can make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] The intelligent curve warning system includes the following levels:

[0037] 1. The perception layer, including sensors and cameras, is used to collect speed data, position data, and vehicle image data of bidirectional vehicles in the curve area; sensors include light sensors and radars.

[0038] 2. The transmission layer includes a network module and a display device. The network module is used to transmit the data collected by the perception layer to the application layer. The display device is used to display the curve warning information; the display device includes an LED display screen and a warning light. The curve warning information includes the license plate number, vehicle type, vehicle distance, and vehicle speed.

[0039] 3. Application layer, used for decoding and classified storage of data.

[0040] 4. The platform layer communicates with the application layer, analyzes the data decoded by the application layer in real time, identifies the vehicle image data information, and sends a curve warning signal to the display device when the data reaches the threshold value.

[0041] 5. The facility layer is a cloud resource module that communicates with the application layer and is used for data storage and cleaning. Data cleaning is implemented using HDOP technology to reduce data errors and obtain more accurate data dimensions.

[0042] A curve intelligent warning method comprises the following steps:

[0043] S1. After the radar in the sensing layer senses that the vehicle enters the curve area, it collects the vehicle speed and position data and triggers the camera to start;

[0044] S2. The camera captures the vehicle and obtains vehicle image data;

[0045] S3. The data collected by radar and camera are transmitted through the transmission layer and application layer to the platform layer, which performs data analysis, algorithm recognition, and data statistics on the data;

[0046] S4. After the platform layer analyzes and identifies the curve warning information, it sends it to the display device for display;

[0047] S5. The warning lights flash in different types according to the distance and speed of the vehicle to warn the vehicle of curves.

[0048] In step S1, the principle of radar sensing the vehicle is:

[0049] S11. Set the range of the radar sector sensing and set the sensing edge distance between 30 meters and 50 meters according to the actual curvature of the curve.

[0050] S12: When a vehicle enters the sensing sector area, the radar effect is triggered, and the vehicle speed and distance information are transmitted back to the control panel of the perception layer.

[0051] The control board receives the data information sent back by the radar, and the direction of its electrical signal is divided into two aspects: on the one hand, the electrical signal is transmitted to the platform layer as a digital signal; on the other hand, the electrical signal is used as a trigger signal and transmitted to the camera through RJ485 or other means.

[0052] In step S2, the camera is divided into a forward sensing shooting module and a backward control module. The backward control module receives the trigger information from the control board and captures the vehicle in the set direction. The captured data is transmitted in two directions: on the one hand, it is transmitted to the platform layer; on the other hand, it is transmitted to the LED display device. In order to solve the problem of camera distortion when capturing vehicles entering a curve at a fast speed, a Bessel-based mathematical method is used to repair the distorted data image, and the repaired image is imported into the deep learning model to train the images of the front and rear capture intervals after processing, and output a complete captured image.

[0053] The main steps are as follows:

[0054] (1) Collect distorted vehicle snapshot image data and select data from it, that is, the vehicle outline is basically intact and the license plate color is clear enough to roughly distinguish its content;

[0055] (2) Establish mathematical methods to process a large amount of data mathematically, and mathematically complete most distorted images, such as vehicle outlines, based on the data method;

[0056] The mathematical method is as follows:

[0057] a. First, according to the image properties, that is, according to the image RGB principle, the distorted image RGB is decomposed into red (R), green (G), and blue (B);

[0058] b. Construct an RGB three-dimensional image, scatter the color points on the three-dimensional coordinates, and construct a three-dimensional coordinate map;

[0059] c. Establish the correlation between the distorted image and the normal image, and use the Pearson correlation coefficient to construct the correlation coefficient between the scattering points on the three-dimensional coordinates.

[0060] The definition is as follows:

[0061]

[0062] The construction introduces both covariance and standard deviation to define the correlation, namely the Pearson correlation coefficient, which is defined as follows:

[0063]

[0064] After simplification, it becomes as follows:

[0065]

[0066] Establish a correlation model and input the correlation parameters into the Bezier curve for smooth repair;

[0067] According to the output data of Bezier curve, a training set is established;

[0068] like Figure 1 As shown in the figure, build an LSTM long short memory training model, import the time series training set into the model for training, and the completed model will correct the distorted image according to the data.

[0069] In step S3, after the captured data is transmitted back to the platform layer, the platform layer uses edge computing and multi-target recognition technology to capture the vehicle information traveling on the curve. It can expose and capture clear vehicle images of vehicles traveling at a speed of 40-60km / h, and identify the vehicle license plate number, vehicle type, and driver's facial information.

[0070] The training of the recognition algorithm consists of the following steps:

[0071] (1) The data training set is captured vehicle images, and the definition set is car (x1, x2, x3....xn);

[0072] (2) Modeling is performed using the conventional Yolo5 visual algorithm + CNN convolutional neural network deep learning algorithm;

[0073] Yolo5 is used to establish a target recognition module and a dynamic vehicle recognition library. As a function of building front-end recognition, other hardware facilities are used to complete the recognition of target vehicles by the front-end camera. The depth of the CNN military aircraft neural network is based on four parts:

[0074] The definition is as follows:

[0075] The core of the convolutional layer is the convolution operation, which uses a set of learnable filters (or convolution kernels) to scan the input data. Suppose there is an input feature map X of size (H, W, C) (height, width, number of channels), and a filter W of size (fh, fw, C), where fh and fw are the height and width of the filter respectively. The convolution operation can be expressed by the following formula to obtain an element of the output feature map O:

[0076]

[0077] Among them, b is the bias term and (i,j) is the position index of the output feature map.

[0078] Construct a coordinate map through a nonlinear activation function;

[0079] f(x) = max(0, x)

[0080] Max pooling is used to reduce the spatial dimensionality and keep the most important features. If a pooling window of size (ph,pw) is used.

[0081]

[0082] Among them, s hand w is the pooling step size;

[0083] After the convolution and pooling layers, there is usually a fully connected layer for classification tasks. If the fully connected layer has D neurons, its input is the flattened vector F of the previous layer (the flattened feature map), the weight matrix is ​​W, and the bias vector is b, then the output of the kth neuron is:

[0084]

[0085] Among them, σ is the activation function.

[0086] (3) Data feature extraction mainly includes vehicle appearance, license plate segmentation, license plate color and face information;

[0087] (4) Extract features from the data training set and label those that need to be identified;

[0088] (5) Output training results, evaluate algorithm model indicators and analyze the recognition accuracy of the algorithm model;

[0089] (6) 2,000 sets of data are continuously input into the algorithm model for training every day.

[0090] The recognition process in step S3 also needs to take into account the overlap caused by multiple vehicles turning in low vision. Conventional recognition is difficult to segment overlapping vehicles in low vision, and thus cannot identify and judge that the objects on the curve are vehicles. Low viewpoint means that the video shooting angle is relatively low and the shooting distance is relatively close, resulting in larger and denser targets, and the scale of the targets changes greatly during movement, accompanied by frequent occlusions. Therefore, the MOT problem of low-viewpoint video shooting, especially the occlusion problem under low viewpoint, will lead to "tracking drift" and "virtual new target" phenomena.

[0091] First, a detection and tracking strategy (TBD) is established. The model trained based on the above recognition algorithm will extract each frame of the data set;

[0092] Construct an algorithm model under adaptive perception and explain some important parameters involved in the algorithm. In the table, the n1 data set is the number of targets that have been successfully detected and tracked from the start frame to the current frame, the n2 data set is the number of all detected targets in the current frame, and f is the current frame number. It can be expressed as:

[0093] n1={1,2,3......i};

[0094] n2={1,2,3......i};

[0095] f={1,2,3......t};

[0096] Obtaining the target appearance features under adaptation

[0097]

[0098] The IOU overlap area between the current frame target and the previous frame target is calculated as the motion feature, and the obtained motion feature is saved in the matrix D.

[0099]

[0100] D i ' ,j =1-D i,j

[0101] According to the mutual coverage degree of the target frames in the current frame, the global occlusion coefficient is calculated, P(k, g) is the total number of occlusions, n2 is the number of targets in the current frame, and η reflects the degree of mutual occlusion of targets in the current frame.

[0102]

[0103] According to the target frame information detected in the current frame, the local occlusion coefficient of each target is calculated, Area(l1 j ) is the area of ​​the target frame detected in the current frame j, and the target local occlusion coefficient It reflects the degree of occlusion of the j-th target in the t-th frame. 0 represents no occlusion, and greater than or equal to 1 represents severe occlusion. The larger the value, the more severe the occlusion.

[0104]

[0105] According to the weights calculated by the local occlusion system, the adaptive mode is used to identify the targets that were not in the low-vision state or were not affected by the target acquisition and the targets in the low-vision state. The adaptiveness is trained according to the occlusion coefficient and weight. Considering that the types of vehicle shapes are relatively limited, a large amount of data training can be avoided in the data set training.

[0106] In step S5, the platform layer can set an upper limit between the warning light and the radar data. The value is associated with the control board through the platform layer, and the sensing radar data reaches the upper limit, triggering the warning light to flash. The warning light is divided into red and blue lights, red light indicates a closer distance, and blue light indicates a farther distance.

[0107] In addition, when there is no vehicle in motion, the warning system has an LED display device that displays text or other broadcast content, and the broadcast content can be customized.

[0108] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to replace some of the technical features therein with equivalents. However, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent curve warning system, characterized by: The perception layer includes sensors and cameras, which are used to collect speed data, position data, and vehicle image data of bidirectional vehicles in the curve area; The transmission layer includes a network module and a display device, wherein the network module is used to transmit the data collected by the perception layer to the application layer, and the display device is used to display the curve warning information; Application layer, used for decoding and classified storage of data; The platform layer communicates with the application layer, analyzes the data decoded by the application layer in real time, identifies the vehicle image data information, and sends a curve warning signal to the display device when the data reaches the threshold value; The facility layer, which is a cloud resource module, communicates with the application layer and is used for data storage and cleaning.

2. The intelligent curve warning system according to claim 1 is characterized by: The sensors include light sensors and radars.

3. The intelligent curve warning system according to claim 2 is characterized in that: The display device comprises an LED display screen and a warning light.

4. The intelligent curve warning system according to claim 3 is characterized by: The curve warning information includes license plate number, vehicle type, vehicle distance, and vehicle speed.

5. The intelligent curve warning method is characterized by: The following steps are involved: S1. After the radar in the sensing layer senses that the vehicle enters the curve area, it collects the vehicle speed and position data and triggers the camera to start; S2. The camera captures the vehicle and obtains vehicle image data; S3. The data collected by radar and camera are transmitted through the transmission layer and application layer to the platform layer, which performs data analysis, algorithm recognition, and data statistics on the data; S4. After the platform layer analyzes and identifies the curve warning information, it sends it to the display device for display; S5. The warning lights flash in different types according to the distance and speed of the vehicle to warn the vehicle of curves.

6. The intelligent curve warning method according to claim 5, characterized in that: In step S3, the data acquired by the platform layer is cleaned by the facility layer.

7. The intelligent curve warning method according to claim 6, characterized in that: The data cleaning adopts Hdop technology.

8. The intelligent curve warning method according to claim 7, characterized in that: In step S3, the platform layer uses edge computing and multi-target recognition technology to capture information about vehicles traveling on curves and identify the license plate number, vehicle type, and driver's facial information.

Images