Unmanned collision early warning system

Abstract

The invention, which belongs to the technical field of unmanned driving, discloses an unmanned driving collision early warning system comprising a feature classifier construction training module, a vehicle acquisition module, a vehicle detection module, a vehicle image size estimation module, an inter-vehicle distance estimation module and an early warning module. According to the method, the vehicle image with the standard size is shot, the focal length of the camera is estimated based on the standard size and the image size of the vehicle, and then the actual distance between the vehicle and the front and rear vehicles is estimated based on the vehicle size in the vehicle image shot in the unmanned driving process, the known included angle between the camera and the horizontal line and the known focal length and height of the camera. Compared with the prior art, the system specially aims at collision early warning of vehicle rear-end collision, the hardware structure is simple, the software algorithm is not complex, and the risk of unmanned rear-end collision can be effectively reduced with low cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned driving, and in particular relates to an unmanned driving collision warning system. Background Art

[0002] With the continuous advancement of technology and the continuous expansion of its applications, driverless technology will profoundly change our travel and lifestyle. Driverless technology refers to the use of on-board sensors, algorithms and computing devices to achieve autonomous driving of vehicles. Safety performance is particularly important. In the field of driverless driving, collision warning systems can ensure driving safety performance. However, large-scale collision warning systems are often very complex, requiring the installation of multiple sensors in all directions of the vehicle and the use of very complex algorithms. The hardware and software requirements are extremely high. Once the hardware or software is damaged or stuck, it will seriously affect the safety of the driverless vehicle.

[0003] During vehicle driving, the applicant discovered that the most common collision is actually rear-end collision. In a complex collision warning system, an independent warning system with relatively simple software and hardware can be developed specifically for rear-end collisions as a supplement to safety protection. Summary of the Invention

[0004] To solve the problems raised in the above background technology, the present invention provides an unmanned driving collision warning system, which is specifically aimed at vehicle rear-end collision warning and can effectively reduce the risk of unmanned driving rear-end collision.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an unmanned driving collision warning system, comprising:

[0006] Feature classifier construction and training module, builds, trains and tests feature classifiers until the preset conditions are met;

[0007] The vehicle acquisition module collects vehicle images in front and behind the vehicle based on the vehicle camera;

[0008] The vehicle detection module detects the vehicle position in the collected vehicle image based on the feature classifier;

[0009] a vehicle image size estimation module, which estimates the image size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position, wherein the image size of the vehicle in the vehicle image includes the height of the vehicle in the vehicle image;

[0010] The inter-vehicle distance estimation module estimates the actual inter-vehicle distance between the vehicle and the vehicles in front and behind based on the known angle between the camera and the horizontal line, the known focal length and height of the camera, and the estimated image size of the vehicle in the vehicle image. The expression is:

[0011]

[0012] Where: θ represents the angle between the camera and the horizontal line, F c Represents the focal length of the camera, H c Represents the height of the camera, y b -y h It is represented as the height of the vehicle in the vehicle image;

[0013] The warning module presets a warning threshold and compares the estimated actual vehicle distance between the vehicle and the front and rear vehicles with the warning threshold. If the estimated vehicle distance is less than the warning threshold, a warning is issued; otherwise, no warning is issued.

[0014] Furthermore, the specific steps of constructing, training and testing the feature classifier in the feature classifier construction and training module include:

[0015] Collect positive sample images containing vehicles and negative sample images that do not contain vehicles;

[0016] The collected images are divided into training dataset and test dataset in a ratio of 7:3;

[0017] Train the feature classifier using the training dataset;

[0018] Use the test dataset to test the trained feature classifier.

[0019] Furthermore, the specific steps of the vehicle image size estimation module estimating the size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position include:

[0020] Selecting feature points of the vehicle in the vehicle image;

[0021] Get the pixel coordinates of the vehicle feature points;

[0022] Calculate pixel distance based on Euclidean distance according to the pixel coordinates of vehicle feature points;

[0023] Calculate the size of the vehicle in the vehicle image based on the shape and pixel distance of the vehicle;

[0024] When the shape formed by the feature points of the vehicle is a rectangle, the formula is:

[0025] S=L×W

[0026] Where: L represents the horizontal pixel distance, and W represents the vertical pixel distance.

[0027] Furthermore, the known angle between the camera and the horizontal line in the vehicle distance estimation module is obtained based on measurements after the camera is installed.

[0028] Furthermore, the known focal length of the camera in the vehicle distance estimation module is obtained based on the following steps:

[0029] Based on camera images of vehicles with known actual dimensions;

[0030] The image size of the vehicle in the vehicle image is estimated based on the pixel coordinates of the vehicle image captured by the camera;

[0031] The camera focal length is estimated based on the known actual size of the vehicle and the image size, as follows:

[0032] 1 / f=1 / v+1 / u

[0033] Where: f represents the focal length of the camera, v represents the image distance, that is, the size of the vehicle in the vehicle image, and u represents the object distance, that is, the known actual size of the vehicle.

[0034] Furthermore, adding positive and negative sample images under various weather and lighting conditions to the training and test datasets can effectively improve the robustness of the system.

[0035] Furthermore, the vehicle detection module detects the position of the vehicle in the collected vehicle image based on the feature classifier, specifically including: first converting the collected vehicle image into a grayscale image, and performing edge detection to obtain the edge contour of the vehicle in the vehicle image, then calculating the symmetry score of each pixel on the edge contour, and selecting the line with the highest score as the symmetry axis; the vehicle image size estimation module calculates the image size of the vehicle in the vehicle image based on the pixel point coordinates and symmetry axis data of the above-mentioned vehicle edge contour.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention captures a vehicle image of known actual size, estimates the camera focal length based on the known actual size of the vehicle and the image size, and then estimates the actual distance between the vehicle and the front and rear vehicles based on the vehicle size in the vehicle image captured during unmanned driving, the known angle between the camera and the horizontal line, and the known focal length and height of the camera. Collision warning is then performed based on the estimated actual distance. Compared with the existing technology, this system is specifically aimed at collision warning of vehicle rear-end collisions, has a simple hardware structure, and a simple software algorithm, and can effectively reduce the risk of unmanned rear-end collisions at a relatively low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a framework diagram of the unmanned collision warning system of the present invention;

[0039] Figure 2 Schematic diagram of vehicle distance estimation of the vehicle distance estimation module of the present invention;

[0040] In the figure: 1. Feature classifier construction training module; 2. Vehicle acquisition module; 3. Vehicle detection module; 4. Vehicle image size estimation module; 5. Vehicle distance estimation module; 6. Early warning module. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] See also Figure 1-2 The present invention provides the following technical solutions: an unmanned driving collision warning system, comprising:

[0043] Feature classifier construction and training module 1: construct, train, and test the feature classifier until it reaches the optimal state. The optimal implementation method here is to preset a conditional threshold in the classifier. For example, when the recognition accuracy reaches 90%, the training is stopped and considered optimal.

[0044] Vehicle acquisition module 2, which uses vehicle cameras to capture images of vehicles in front and behind the vehicle. The vehicle cameras can be single cameras installed at the front and rear of the vehicle, or multiple cameras can be used. To save hardware costs, a monocular camera can be installed approximately in the center of the front and rear of the vehicle to capture images of other vehicles in front and behind the vehicle;

[0045] The vehicle detection module 3 detects the vehicle position in the collected vehicle image based on the feature classifier;

[0046] A vehicle image size estimation module 4 estimates the image size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position, wherein the image size of the vehicle in the vehicle image includes the height of the vehicle in the vehicle image;

[0047] The inter-vehicle distance estimation module 5 estimates the actual inter-vehicle distance between the vehicle and the vehicles in front and behind based on the known angle between the camera and the horizontal line, the known focal length and height of the camera, and the estimated image size of the vehicle in the vehicle image. The expression is:

[0048]

[0049] Where: θ represents the angle between the camera and the horizontal line, F c Represents the focal length of the camera, H c Represents the height of the camera, y b -y hIt is represented as the height of the vehicle in the vehicle image;

[0050] The warning module 6 presets a warning threshold, compares the estimated actual vehicle distance between the vehicle and the front and rear vehicles with the warning threshold, and issues a warning if the estimated vehicle distance is less than the warning threshold, otherwise, no warning is issued.

[0051] Specifically, the specific steps of constructing, training, and testing the feature classifier in the feature classifier construction and training module 1 include:

[0052] Collect positive sample images containing vehicles and negative sample images that do not contain vehicles;

[0053] Positive and negative sample images include images under various weather and lighting conditions to enhance the robustness of the feature classifier;

[0054] The collected images are divided into training dataset and test dataset in a ratio of 7:3;

[0055] Train the feature classifier using the training dataset;

[0056] The feature classifier can be a classifier that detects vehicle features, or a classifier that detects a combination of weather and vehicle features. Including positive and negative sample images under various weather and lighting conditions in the training and test datasets can effectively increase the robustness of the system. The first classifier directly detects vehicle features in vehicle images, while the second classifier first detects weather conditions in vehicle images and then performs targeted vehicle feature detection based on the detected weather conditions, including adjusting the vehicle feature detection threshold. The vehicle feature detection threshold corresponding to the weather conditions is the optimal parameter pre-set during the training process.

[0057] Use the test dataset to test the trained feature classifier.

[0058] Specifically, the vehicle detection module (3) detects the position of the vehicle in the collected vehicle image based on the feature classifier, specifically including: first converting the collected vehicle image into a grayscale image, and performing edge detection to obtain the edge contour of the vehicle in the vehicle image, then calculating the symmetry score of each pixel on the edge contour, and selecting the line with the highest score as the symmetry axis; the vehicle image size estimation module (4) calculates the image size of the vehicle in the vehicle image based on the pixel coordinates of the edge contour of the vehicle and the symmetry axis data.

[0059] Specifically, the vehicle image size estimation module 4 estimates the size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position, including the following steps:

[0060] Selecting feature points of the vehicle in the vehicle image;

[0061] Get the pixel coordinates of the vehicle feature points;

[0062] Calculate pixel distance based on Euclidean distance according to the pixel coordinates of vehicle feature points;

[0063] Calculate the size of the vehicle in the vehicle image based on the shape of the vehicle, pixel distance and formula;

[0064] When the shape formed by the feature points of the vehicle is a rectangle, the formula is:

[0065] S=L×W

[0066] Where: L represents the horizontal pixel distance, and W represents the vertical pixel distance.

[0067] When the shape formed by the feature points of a vehicle is non-rectangular, it is necessary to perform matrix correction on the non-rectangular frame before calculating the vehicle image size.

[0068] Specifically, the known angle between the camera and the horizontal line in the vehicle distance estimation module 5 is obtained based on measurements after the camera is installed.

[0069] Specifically, the camera focal length in the vehicle distance estimation module 5 is known based on the following steps:

[0070] A vehicle with known actual size based on camera capture can also be called a standard size vehicle;

[0071] The image size of the vehicle in the vehicle image is estimated based on the pixel coordinates of the vehicle image captured by the camera;

[0072] The camera focal length is estimated based on the known actual size of the vehicle and the image size, as follows:

[0073] 1 / f=1 / v+1 / u

[0074] Where: f represents the focal length of the camera, v represents the image distance, that is, the size of the vehicle in the vehicle image, and u represents the object distance, that is, the known actual size of the vehicle.

[0075] The present invention captures a vehicle image of known actual size, estimates the camera focal length based on the known actual size of the vehicle and the image size, and then estimates the actual distance between the vehicle and the front and rear vehicles based on the vehicle size in the vehicle image captured during unmanned driving, the known angle between the camera and the horizontal line, and the known focal length and height of the camera. Collision warning is then performed based on the estimated actual distance. Compared with the existing technology, this system is specifically aimed at collision warning of vehicle rear-end collisions, has a simple hardware structure, and a simple software algorithm, and can effectively reduce the risk of unmanned rear-end collisions at a relatively low cost.

[0076] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An unmanned driving collision warning system, characterized in that: include: Feature classifier construction and training module (1) constructs, trains and tests the feature classifier until the preset conditions are met; A vehicle acquisition module (2) acquires vehicle images in front of and behind the vehicle based on a vehicle camera; A vehicle detection module (3) detects the position of a vehicle in the collected vehicle image based on a feature classifier; A vehicle image size estimation module (4) estimates the image size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position, wherein the image size of the vehicle in the vehicle image includes the height of the vehicle in the vehicle image; The vehicle distance estimation module (5) estimates the actual vehicle distance between the vehicle and the front and rear vehicles based on the known angle between the camera and the horizontal line, the known focal length and height of the camera, and the estimated image size of the vehicle in the vehicle image. The expression is: Where: θ represents the angle between the camera and the horizontal line, F c Represents the focal length of the camera, H c Represents the height of the camera, y b -y h It is represented as the height of the vehicle in the vehicle image; An early warning module (6) presets an early warning threshold, compares the estimated actual vehicle distance between the vehicle and the front and rear vehicles with the early warning threshold, and issues an early warning if the estimated vehicle distance is less than the early warning threshold; Otherwise, no warning will be issued.

2. The unmanned driving collision warning system according to claim 1, characterized in that: The specific steps of constructing, training and testing the feature classifier in the feature classifier construction and training module (1) include: Collect positive sample images containing vehicles and negative sample images that do not contain vehicles; The collected images are divided into training dataset and test dataset in a ratio of 7:3; Train the feature classifier using the training dataset; Use the test dataset to test the trained feature classifier.

3. The unmanned driving collision warning system according to claim 1, characterized in that: The specific steps of the vehicle image size estimation module (4) estimating the size of the vehicle in the vehicle image based on the image pixel coordinates of the detected vehicle position include: Selecting feature points of the vehicle in the vehicle image; Get the pixel coordinates of the vehicle feature points; Calculate pixel distance based on Euclidean distance according to the pixel coordinates of vehicle feature points; Calculate the size of the vehicle in the vehicle image based on the shape and pixel distance of the vehicle; When the shape formed by the feature points of the vehicle is a rectangle, the formula is: S=L×W Where: L represents the horizontal pixel distance, and W represents the vertical pixel distance.

4. The unmanned driving collision warning system according to claim 1, characterized in that: The known angle between the camera and the horizontal line in the vehicle distance estimation module (5) is obtained based on measurements after the camera is installed.

5. The unmanned driving collision warning system according to claim 1, characterized in that: The focal length of the camera in the vehicle distance estimation module (5) is known based on the following steps: Based on camera images of vehicles with known actual dimensions; The image size of the vehicle in the vehicle image is estimated based on the pixel coordinates of the vehicle image captured by the camera; The camera focal length is estimated based on the known actual size of the vehicle and the image size, as follows: 1 / f=1 / v+1 / u Where: f represents the focal length of the camera, v represents the image distance, that is, the image size of the vehicle in the vehicle image, and u represents the object distance, that is, the known actual size of the vehicle.

6. The unmanned driving collision warning system according to claim 2, characterized in that: Positive and negative sample images under various weather and lighting conditions are added to the training and test datasets.

7. The unmanned driving collision warning system according to any one of claims 1 to 6, characterized in that: The vehicle detection module (3) detects the position of the vehicle in the collected vehicle image based on the feature classifier, specifically comprising: first converting the collected vehicle image into a grayscale image, performing edge detection to obtain the edge contour of the vehicle in the vehicle image, then calculating the symmetry score of each pixel on the edge contour, and selecting the line with the highest score as the symmetry axis; The vehicle image size estimation module (4) calculates the image size of the vehicle in the vehicle image based on the pixel coordinates and symmetry axis data of the edge contour of the vehicle.

Images