Automobile anti-collision warning method, system, device and medium based on machine vision

Abstract

The present invention relates to the field of automobile monitoring systems, and specifically to an automobile anti-collision warning method, system, device and medium based on machine vision, including: using multiple high-precision cameras to obtain images of the vehicle's surrounding environment; estimating the estimated position of moving objects around the vehicle based on the environmental image; correcting the estimated position of the moving object through a ranging radar to obtain a corrected position; predicting the moving trajectory of the moving object based on the corrected position extracted from continuous video frames; judging the collision probability based on the future moving trajectory and the vehicle trajectory; and sending a warning signal to the driver through at least one warning channel according to the collision probability. This helps to improve driving safety and reduce the incidence of traffic accidents.

Description

Technical Field

[0001] The present invention relates to the field of automobile monitoring systems, and in particular to a machine vision-based automobile anti-collision warning method, system, device and medium. Background Art

[0002] The problem of automobile road collision refers to the collision between vehicles or between vehicles and other obstacles (such as pedestrians, bicycles, motorcycles, etc.) due to various reasons such as improper driver operation, vehicle failure, poor road conditions, etc. during the driving process. These collision accidents may cause casualties, vehicle damage and property loss, and even cause more serious accidents such as multi-vehicle chain collisions and fires. In order to reduce the occurrence of collision accidents, relevant departments and institutions have been working hard to strengthen road traffic safety management, improve vehicle safety performance, enhance driver safety awareness, and strengthen traffic safety publicity and education. At the same time, automobile safety collision tests are also an important means of evaluating vehicle safety performance. Through strict collision test standards, it can be ensured that the vehicle can provide sufficient protection measures in the event of a collision and reduce the damage to the occupants caused by the collision accident.

[0003] Existing car collisions are mainly judged by a single radar, which has low accuracy and great limitations. Summary of the invention

[0004] The purpose of the present invention is to provide a machine vision-based automobile anti-collision warning method, system, device and medium, aiming to use cameras and radars to comprehensively perform anti-collision warning judgments, thereby improving driving safety and reducing the incidence of traffic accidents.

[0005] To achieve the above-mentioned object, in a first aspect, the present invention provides a vehicle anti-collision warning method based on machine vision, comprising using multiple high-precision cameras to acquire images of the vehicle's surrounding environment;

[0006] estimating the estimated position of the moving objects around the vehicle based on the environment image;

[0007] The estimated position of the moving object is corrected by using the ranging radar to obtain a corrected position;

[0008] Predict the trajectory of a moving object based on the corrected position extracted from consecutive video frames;

[0009] Determine the collision probability based on the future movement trajectory and vehicle trajectory;

[0010] A warning signal is sent to the driver through at least one warning channel according to the collision probability.

[0011] The specific steps of using multiple high-precision cameras to acquire images of the vehicle's surrounding environment include:

[0012] Continuously capture surrounding environment images through multiple high-precision cameras;

[0013] transmitting the captured environment image from the camera to the central processing unit;

[0014] The environment image is preprocessed by the central processing unit.

[0015] The specific steps of estimating the estimated position of the moving object around the vehicle based on the environment image include:

[0016] Apply the YOLO object detection algorithm to identify moving objects in the preprocessed environment image;

[0017] Calculate the center point of the moving object and generate its position in the image;

[0018] The depth information of the moving object is estimated by multi-view geometric fusion of multiple high-precision cameras, thereby obtaining the estimated position of the moving object in three-dimensional space.

[0019] The specific steps of correcting the estimated position of the moving object by using the ranging radar to obtain the corrected position include:

[0020] Determine the position of the moving object detected by the radar in the radar coordinate system;

[0021] Converting the position data in the radar coordinate system into coordinates corresponding to the image coordinate system captured by the vehicle camera;

[0022] Compare the position data provided by the radar with the position data estimated based on the environment image;

[0023] The estimated position is corrected according to the comparison result to obtain the corrected position.

[0024] The specific steps of predicting the moving trajectory of the moving object based on the corrected position extracted from continuous video frames include:

[0025] Using the position data in consecutive frames, the trajectory of the moving object is constructed;

[0026] The Kalman filter is applied based on the trajectory of the moving object to predict the future movement trajectory of the moving object.

[0027] The specific steps of applying the Kalman filter to predict the future movement trajectory of the moving object based on the trajectory of the moving object include:

[0028] Choose two-dimensional variables to describe the position and velocity of the moving object;

[0029] Setting the state transfer equation and observation model under the condition that the acceleration is constant during a preset evaluation period;

[0030] Initialize the Kalman filter based on the initial observation data;

[0031] Use the observation model to predict the current state based on the state estimate at the previous moment.

[0032] The specific steps of determining the probability based on the future moving trajectory and the vehicle trajectory include:

[0033] Calculate the minimum distance and time interval between the two trajectories based on the future moving trajectory and the vehicle trajectory;

[0034] If a potential collision is detected, the probability of collision is assessed based on the number of possible avoidance measures;

[0035] Risks are graded according to the probability of collision.

[0036] In a second aspect, the present invention provides a vehicle anti-collision warning system based on machine vision, comprising: an image data acquisition module, a position estimation module, a radar correction module, a movement trajectory prediction module, a collision probability judgment module and a warning module;

[0037] The image data acquisition module is used to acquire images of the vehicle's surrounding environment using multiple high-precision cameras;

[0038] The position estimation module is used to estimate the estimated position of the moving objects around the vehicle based on the environment image;

[0039] The radar correction module is used to correct the estimated position of the moving object by using the ranging radar to obtain a corrected position;

[0040] The movement trajectory prediction module is used to predict the movement trajectory of the moving object based on the corrected position extracted from the continuous video frames;

[0041] The collision probability judgment module is used to judge the collision probability based on the future movement trajectory and the vehicle trajectory;

[0042] The warning module is used to send a warning signal to the driver through at least one warning channel according to the collision probability.

[0043] In a third aspect, the present invention provides a vehicle anti-collision warning device based on machine vision, comprising a processor, an interface and a memory, wherein the processor and the interface are connected to the memory and a vehicle expansion port, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the vehicle anti-collision warning method based on machine vision.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the machine vision-based automobile anti-collision warning method.

[0045] The machine vision-based automobile anti-collision warning method, system, device and medium of the present invention first use multiple high-precision cameras to obtain images of the vehicle's surrounding environment. These cameras are distributed in different positions and can cover a 360-degree viewing angle around the vehicle to ensure comprehensive monitoring of the surrounding environment. The images captured by the camera are used to identify moving objects around the vehicle through image processing technology, such as edge detection, target recognition, etc. Next, the estimated position of the moving object around the vehicle is estimated based on the environmental image. This step uses computer vision technology, such as optical tracking, Kalman filtering, etc., to model the motion state of the moving object and predict its future position. In addition, deep learning methods, such as convolutional neural networks (CNN) and recurrent neural networks (RNN), etc., can also be used to improve the accuracy of the estimated position. Then, the estimated position of the moving object is corrected by a ranging radar to obtain a corrected position. The ranging radar calculates the distance between the target object and the vehicle by emitting radio waves and receiving reflected signals, thereby accurately correcting the estimated position. The corrected position is more accurate and helps in subsequent collision warning judgments. Based on the corrected position extracted from continuous video frames, the moving trajectory of the moving object can be predicted. This step predicts the future movement trajectory by analyzing the motion state of the moving object in the continuous video frames, combining the estimated position and the corrected position. At the same time, the movement trajectory is updated in real time to adapt to the movement changes of the object. Next, the collision probability is judged based on the future movement trajectory and the vehicle trajectory. This step calculates the possibility of collision by comparing the trajectory of the moving object and the vehicle. The judgment methods include distance judgment, speed judgment, etc., and the appropriate judgment criteria can be selected according to the actual situation. When the collision probability exceeds the set threshold, it is considered that there is a collision risk. Finally, according to the collision probability, a warning signal is sent to the driver through at least one warning channel. The warning channels include sound, light, vibration, etc., which can be selected according to actual needs and driver habits. The intensity and form of the warning signal can be adjusted according to the size of the collision probability to improve the driver's alertness. This helps to improve driving safety and reduce the incidence of traffic accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 It is a flow chart of the automobile anti-collision warning method based on machine vision according to the first embodiment of the present invention.

[0048] Figure 2 It is a flow chart of the first embodiment of the present invention for acquiring an image of the vehicle's surrounding environment using multiple high-precision cameras.

[0049] Figure 3 is a flowchart of estimating the estimated position of a moving object around a vehicle based on the environment image according to the first embodiment of the present invention.

[0050] Figure 4 This is a flow chart of the first embodiment of the present invention for correcting the estimated position of a mobile object by using a ranging radar to obtain a corrected position.

[0051] Figure 5 It is a flowchart of predicting the moving trajectory of a moving object based on the corrected position extracted from continuous video frames according to the first embodiment of the present invention.

[0052] Figure 6 The flowchart of the first embodiment of the present invention is to predict the future movement trajectory of the moving object by applying the Kalman filter based on the movement trajectory of the moving object.

[0053] Figure 7 This is a flow chart of determining the collision probability based on the future moving trajectory and the vehicle trajectory according to the first embodiment of the present invention.

[0054] Figure 8 This is a flowchart of evaluating the collision probability based on the number of avoidance measures that can be taken if a potential collision is detected according to the first embodiment of the present invention.

[0055] Fig. 9 It is a flow chart of sending a warning signal to a driver through at least one warning channel according to the collision probability according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0056] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0057] First embodiment

[0058] See also Figures 1 to 9 The present invention provides a vehicle anti-collision warning method based on machine vision, comprising:

[0059] S101 uses multiple high-precision cameras to acquire images of the vehicle’s surroundings;

[0060] The specific steps include:

[0061] S201 continuously captures images of the surrounding environment through multiple high-precision cameras;

[0062] Multiple high-precision cameras continuously capture images of the environment around the vehicle. These cameras are distributed around the entire vehicle to ensure that every corner around the vehicle is covered, providing comprehensive data support for subsequent processing.

[0063] S202 transmits the captured environment image from the camera to the central processing unit;

[0064] The captured environment image is transmitted from the camera to the central processing unit. This process requires efficient data transmission and processing capabilities because the number of images captured per second is huge. The delay and data loss during the transmission process must be minimized to ensure the integrity and real-time nature of the image.

[0065] S203 pre-processes the environment image through the central processing unit.

[0066] The central processing unit will preprocess the environmental image. Preprocessing includes noise removal, color balance, and contrast enhancement operations to improve the quality of the image. In the field of computer vision, the central processing unit (CPU) often needs to perform a series of preprocessing operations when processing environmental images. This is because environmental images may be affected by various factors such as noise, color imbalance, and lack of contrast during acquisition, transmission, and storage. In order to improve the quality of the image and ensure the accuracy of subsequent processing and analysis, the central processing unit will preprocess the image.

[0067] Image preprocessing mainly includes the following aspects.

[0068] 1. Noise removal: Noise is a common problem in the process of image acquisition and transmission. Noise may come from sensors, transmission lines or other external interference. The purpose of noise removal is to eliminate these interferences and improve the clarity and contrast of the image. Common noise removal methods include mean filtering, median filtering, and Gaussian filtering.

[0069] 2. Color balance: In environmental images, the color of objects may be affected by lighting conditions, shooting angles, and object materials, resulting in color distortion. Color balance is to adjust the color of the image to make it closer to the real color. The color balance methods include histogram equalization, adaptive histogram equalization, and gamma transformation.

[0070] 3. Contrast enhancement: Contrast is a measure of the difference in pixel values ​​in an image. Contrast enhancement is to make the target objects in the image clearer and the edges sharper. There are many contrast enhancement methods, such as linear transformation, piecewise linear transformation, and histogram equalization.

[0071] 4. Image sharpening: Image sharpening is to improve the clarity and edge sharpness of the target object in the image. Common sharpening methods include Gaussian sharpening, unsharp masking, and guided filtering.

[0072] S102 estimates the estimated position of the moving objects around the vehicle based on the environment image;

[0073] The specific steps include:

[0074] S301 applies the YOLO target detection algorithm to identify moving objects in the preprocessed environment image;

[0075] In the target detection stage, the YOLO (You Only Look Once) algorithm is used to identify moving objects in the preprocessed environment image. The YOLO algorithm is a real-time target detection method with the characteristics of fast detection speed and high accuracy. In this stage, the preprocessed image is first segmented, and then the target category prediction and bounding box positioning are performed for each segmented area, so as to accurately identify the moving objects in the image. In order to achieve this goal, the preprocessed image needs to be effectively segmented first, and the entire image needs to be divided into multiple regions with different characteristics. Next, the target category prediction and bounding box positioning need to be performed for each segmented area, so as to accurately identify the moving objects in the image. There are many methods used in image segmentation, such as threshold-based methods, region growing-based methods, and edge detection-based methods. These methods have their own advantages and disadvantages and need to be selected according to the actual application scenarios and needs. For example, the threshold-based method works better when processing images with obvious color differences, while the region growing and edge detection-based methods have advantages in processing image segmentation in complex scenes. After completing the image segmentation, the target category prediction needs to be performed for each segmented area. This step usually uses deep learning methods, such as convolutional neural networks (CNN). By extracting features and classifying the segmented regions, the target category to which each region belongs is obtained. It should be noted that when making category predictions, appropriate loss functions and optimization algorithms should be selected to improve the prediction accuracy. Next, the bounding box of each segmented region needs to be located. The accuracy of bounding box positioning directly affects the final recognition result of the moving object. In order to improve the positioning accuracy, deep learning-based target detection methods such as YOLO and Faster R-CNN are used. These methods can accurately locate the target bounding box while maintaining high recognition accuracy. Finally, the prediction results are integrated and post-processed to accurately identify the moving objects in the image. In the post-processing stage, the non-maximum suppression (NMS) method is used to eliminate redundant box selection results and retain the most likely moving objects. In addition, the recognition results are further optimized by setting thresholds and adjusting box selection parameters.

[0076] S302 calculates the center point of the moving object and generates its position in the image;

[0077] Get the bounding box of the object. The coordinates of the bounding box are the location of the object in the image. Based on this, calculate the center point of the object.

[0078] The calculated coordinates of the center point of the moving object are mapped to the image coordinate system to obtain the position of the center point in the image. It should be noted that there is a certain conversion relationship between the image coordinate system and the actual world coordinate system, which needs to be adjusted according to the actual situation.

[0079] The transformation relationship from the actual world coordinate system to the image coordinate system mainly includes the following aspects:

[0080] 1. Camera intrinsic parameters: The camera intrinsic parameters include focal length, principal point coordinates, and pixel size parameters. These parameters determine the distance in the real world corresponding to a pixel in the image. The determination of camera intrinsic parameters usually requires camera calibration methods.

[0081] 2. Camera extrinsics: Camera extrinsics mainly include the camera’s pose (translation and rotation) matrix. These extrinsics describe the position and direction of the camera in the real world. When calculating the center point of a moving object, we need to synchronize the camera extrinsics with the real world coordinate system to ensure that the object’s motion trajectory is accurately captured.

[0082] 3. Object motion model: In order to convert the object motion in the real world coordinate system into the image coordinate system, we need to establish an object motion model. This model describes the acceleration, velocity, and position information of the object in the real world and maps it to the image coordinate system. Common motion models include constant velocity model, constant acceleration model, and polynomial model.

[0083] 4. Image processing method: After mapping the object motion in the real world coordinate system to the image coordinate system, we need to use certain image processing methods to extract the center point of the object. This includes edge detection, corner detection, and morphological processing methods. According to different application scenarios, choose the appropriate image processing algorithm to complete the positioning of the center point.

[0084] S303 estimates the depth information of the moving object by geometrically fusing multiple views of multiple high-precision cameras, thereby obtaining an estimated position of the moving object in three-dimensional space.

[0085] Multi-view geometric fusion technology integrates image information from multiple perspectives to generate a complete three-dimensional model of a moving object. This process mainly includes image correction, feature extraction, matching and fusion steps. Image correction is to eliminate the distortion caused by the camera and ensure the accuracy of the image. Feature extraction and matching are to find the association between images from different perspectives, thereby providing a basis for fusion. Finally, the fusion step integrates the information from multiple perspectives to obtain the precise position of the moving object in three-dimensional space. The process of multi-view geometric fusion technology is divided into four main steps: image correction, feature extraction, matching and fusion. The first is image correction. The main purpose of this step is to eliminate the distortion caused by the camera to ensure the accuracy of the image. During the shooting process of the camera, due to factors such as the lens and sensor, the image will be distorted, affecting the effect of subsequent processing. Therefore, image correction becomes the key first step of multi-view geometric fusion technology. After completing image correction, the next step is feature extraction and matching. Feature extraction is to extract points or areas with unique properties from the image, such as corners and edges, so as to facilitate matching between images from different perspectives. Feature matching is to find the association between images from different perspectives and find matching feature point pairs, thereby providing a basis for subsequent fusion. This step is the core link in the multi-view geometric fusion technology and affects the quality of the final fusion effect. Finally, the fusion step integrates the information of multiple perspectives to obtain the precise position of the moving object in three-dimensional space. This step is usually implemented using geometry and optimization algorithms, such as triangulation and PnP (Perspective-n-Point) problem solving.

[0086] S103 corrects the estimated position of the moving object by using the ranging radar to obtain a corrected position;

[0087] The specific steps include:

[0088] S401 determines the position of the moving object detected by the radar in the radar coordinate system;

[0089] Radar transmits radio waves, receives reflected signals, and then analyzes these signals to obtain the position and speed information of the target. In the radar coordinate system, the position of a moving object is determined by measuring the time difference of the reflected signal. In order to achieve accurate positioning of moving objects, the distance resolution of the radar transmitting and receiving devices is improved to more accurately identify the distance change between the target object and the radar. We need to understand the concept of radar range resolution. Radar range resolution refers to the minimum distance at which the radar system can distinguish adjacent target objects. Improving the range resolution means that the radar can measure the distance of the target object more accurately, thereby improving the recognition accuracy.

[0090] There are several aspects to improving the range resolution of radar transmitting and receiving devices:

[0091] 1. Improve the design of radar transmitting and receiving devices: By adopting higher transmission power, narrower pulse width and more advanced receiver technology, the strength and sensitivity of radar signals can be improved, thereby improving distance resolution.

[0092] 2. Optimize signal processing methods: Use more efficient signal processing algorithms, such as pulse compression and spectrum analysis, to improve the resolution of radar echoes. By improving the distance resolution, the position, speed and shape of the target object can be judged more accurately, providing more reliable information for various application scenarios.

[0093] S402 converts the position data in the radar coordinate system into coordinates corresponding to the image coordinate system captured by the vehicle camera;

[0094] Converting the position data in the radar coordinate system to the coordinates corresponding to the image coordinate system captured by the vehicle camera is a key step in autonomous driving technology. This process involves the conversion between two major coordinate systems in order to fuse the position information detected by the radar with the image information captured by the camera, thus providing more accurate perception capabilities for the vehicle's autonomous driving.

[0095] To convert the position data in the radar coordinate system into the image coordinate system, the following steps need to be completed:

[0096] 1. Determine the transformation relationship: According to the relative position and posture between the radar and the camera, establish the coordinate system transformation relationship between the two. This transformation relationship usually includes two parts: translation and rotation, which are used to describe the transformation process from the radar coordinate system to the image coordinate system.

[0097] 2. Data fusion: After the coordinate system conversion is completed, the position information detected by the radar needs to be fused with the image information captured by the camera. This is achieved by interpolating and fusing the position data in the radar coordinate system with the pixel coordinates in the image coordinate system. The result of data fusion will provide more comprehensive and accurate perception information for the autonomous driving system.

[0098] 3. Optimization and verification: In order to ensure the accuracy and stability of the conversion results, the conversion process needs to be optimized and verified. This includes the correction of the conversion relationship, real-time monitoring of the fusion data, and adjustment of the system according to the actual situation.

[0099] S403 compares the position data provided by the radar with the position data estimated based on the environment image;

[0100] The position data provided by the radar is compared with the position data estimated by the environmental image, and the difference between the two is analyzed. In order to improve the accuracy of the position estimation, a data fusion algorithm (such as Kalman filter, particle filter) is used to fuse the two data to obtain a more reliable position estimation result.

[0101] S404 corrects the estimated position according to the comparison result to obtain a corrected position.

[0102] Correction is performed based on the error value of the comparison result to obtain the corrected position.

[0103] S104 predicts the moving trajectory of the moving object based on the corrected position extracted from the continuous video frames;

[0104] The specific steps include:

[0105] S501 constructs a moving object trajectory using position data in consecutive frames;

[0106] The position data of the moving object is obtained in the continuous video frames. This data is used to construct the trajectory of the moving object. By analyzing the position data in the continuous frames, the changing rules of the moving object in space and time are understood. This step is the basis for predicting the trajectory of the moving object and provides key information for subsequent predictions.

[0107] S502 predicts the future movement trajectory of the moving object by applying a Kalman filter based on the trajectory of the moving object.

[0108] Based on the construction of the moving object trajectory, the Kalman filter technology is applied to predict the future moving trajectory of the moving object. The Kalman filter is a linear optimal estimation algorithm suitable for processing real-time measurement data containing noise. By using the Kalman filter to predict the future position of the moving object, it provides a basis for other related applications. The specific steps include:

[0109] S601 selects a two-dimensional variable to describe the position and speed of the moving object;

[0110] S602 sets a state transfer equation and an observation model when the acceleration is constant within a preset evaluation time period;

[0111] Before applying the Kalman filter for prediction, it is necessary to set the state transfer equation and observation model. The state transfer equation describes the process of the system transferring from the current state to the next state, while the observation model represents the observed value of the system state. In this process, it is assumed that the acceleration of the moving object is constant, which simplifies the prediction process.

[0112] S603 initializes the Kalman filter according to the initial observation data;

[0113] This step is to ensure that the Kalman filter can correctly process the input data and output the prediction results. Initializing the Kalman filter includes setting the initial state estimate and covariance matrix parameters.

[0114] S604 uses the observation model to predict the state at the current moment based on the state estimation at the previous moment.

[0115] After using the observation model, we predict the state of the mobile object at the current moment based on the state estimate of the previous moment. This step is based on the core function of the Kalman filter, which provides us with real-time information about the position of the mobile object. The state of the mobile object is continuously updated by analyzing consecutive frames, so as to predict its future trajectory more accurately.

[0116] S105 determines the collision probability based on the future movement trajectory and the vehicle trajectory;

[0117] The specific steps include:

[0118] S701 calculates the minimum distance and time interval between the two trajectories based on the future moving trajectory and the vehicle trajectory;

[0119] Specifically, the trajectory can be divided into several small segments, and then the minimum distance and time interval between each small segment are calculated. Finally, the minimum distance and time interval of each small segment are combined to obtain the minimum distance and time interval of the entire trajectory.

[0120] S702 If a potential collision is detected, assess the collision probability based on the number of avoidance measures that can be taken;

[0121] The specific steps include:

[0122] S801 lists all possible risk avoidance measures, including changing speed, changing direction, and performing emergency braking.

[0123] The system lists all possible evasive measures, including changing speed, changing direction, and performing emergency braking. These evasive measures provide the driver with ways to avoid the collision.

[0124] S802: correcting the feasibility of each risk avoidance measure according to the driving state of the vehicle to obtain the number of risk avoidance measures that can be taken;

[0125] The system will correct the feasibility of each risk avoidance measure according to the driving state of the vehicle. The purpose of this step is to take into account that in actual situations, some risk avoidance measures may not be implemented under certain driving conditions. By correcting the feasibility of risk avoidance measures, the risk avoidance measures that the driver can take can be evaluated more accurately.

[0126] S803 calculates the collision probability according to the ratio of the corrected number of risk avoidance measures that can be taken to the total risk avoidance measures.

[0127] The system will calculate the collision probability based on the ratio of the corrected number of possible avoidance measures to the total avoidance measures. This ratio reflects the driver's probability of successfully avoiding a collision when encountering a collision risk. Through this step, the driver is provided with an early warning of the collision risk, which helps to improve driving safety.

[0128] S703 classifies the risks according to the magnitude of the collision probability.

[0129] This method classifies vehicles' collision risk into four levels: very low, low, medium and high.

[0130] In modern transportation, the collision risk assessment of vehicles is particularly important. In order to better manage and control traffic accidents, my country has adopted a new method to classify vehicle collision risks. This method divides vehicle collision risks into four levels: very low, low, medium and high. Below we will introduce in detail the meaning of these four levels and how to assess and manage vehicle collision risks based on these levels.

[0131] First, let's take a look at the very low collision risk level. In this level, the collision risk of the vehicle is almost negligible. This means that the possibility of an accident while the vehicle is driving is very small. Next is the low collision risk level. In this level, the collision risk of the vehicle is relatively low. This means that the possibility of an accident while the vehicle is driving is small. Then there is the medium collision risk level. In this level, the collision risk of the vehicle is at a medium level. This means that the possibility of an accident while the vehicle is driving is high. Finally, there is the high collision risk level. In this level, the collision risk of the vehicle is very high.

[0132] S106 sends a warning signal to the driver through at least one warning channel according to the collision probability.

[0133] In the field of autonomous driving, the importance of the early warning system is self-evident. When the probability of collision is high, the early warning system can send a warning signal to the driver through at least one early warning channel, thereby reminding the driver to pay attention and reduce the occurrence of traffic accidents. The specific steps include:

[0134] S901 When the collision probability exceeds a set threshold, a warning signal is generated;

[0135] S902 sends a warning signal to the driver through the selected warning channel;

[0136] Next, the warning signal will be sent to the driver through the selected warning channel. This warning channel can be sound, vision or other sensory methods, such as vibration. This is to ensure that the driver can effectively receive the warning signal under various driving conditions.

[0137] S903 monitors the driver's feedback to the warning signal within the set reaction time, and continues to warn if there is no feedback.

[0138] During the set reaction time, the warning system will continue to monitor the driver's response to the warning signal. This reaction time is to give the driver enough reaction space to avoid panic operation in an emergency. If the driver does not give feedback within this time, the system will continue to send warning signals until the driver responds.

[0139] This warning method can greatly improve the safety of drivers and reduce the occurrence of traffic accidents. However, it also requires the driver to be sensitive to the warning signal and respond in time.

[0140] Second embodiment

[0141] The present invention also provides a machine vision-based automobile anti-collision warning system, comprising: an image data acquisition module, a position estimation module, a radar correction module, a movement trajectory prediction module, a collision probability judgment module and a warning module; the image data acquisition module is used to use multiple high-precision cameras to obtain an image of the vehicle's surrounding environment; the position estimation module is used to estimate the estimated position of a moving object around the vehicle based on the environmental image; the radar correction module is used to correct the estimated position of the moving object by using a ranging radar to obtain a corrected position; the movement trajectory prediction module is used to predict the movement trajectory of the moving object based on the corrected position extracted from continuous video frames; the collision probability judgment module is used to judge the collision probability based on the future movement trajectory and the vehicle trajectory; the warning module is used to send a warning signal to the driver through at least one warning channel according to the collision probability.

[0142] In this embodiment, the present application provides a vehicle anti-collision warning system based on machine vision. The system integrates multiple modules to achieve real-time monitoring of the vehicle's surrounding environment, thereby effectively improving driving safety.

[0143] The image data acquisition module is the core part of the system. It uses multiple high-precision cameras to capture environmental images around the vehicle. These images will provide data support for subsequent modules. Through high-definition images, moving objects around the vehicle can be more accurately identified, providing an important basis for collision prevention. After receiving the image data, the position estimation module estimates the position of the moving object. This module uses advanced image processing technologies such as object recognition and deep learning to locate the moving object in the image. The accuracy of the estimated position will directly affect the performance of the early warning system, so a variety of algorithms and technologies are used to ensure the accuracy of the position estimation. The radar correction module is to improve the accuracy of the position estimation. It corrects the estimated position of the moving object through the ranging radar to obtain a more accurate corrected position. Radar technology has high accuracy in measuring distance and speed, and combined with image data, it effectively improves the system's estimation of the position of the moving object. On this basis, the moving trajectory prediction module comes into play. It predicts the moving trajectory of the moving object based on the corrected position extracted from continuous video frames. By tracking the trajectory of the moving object, the potential collision risk can be predicted in advance, providing data support for the early warning module. The collision probability judgment module is a key part in assessing the collision risk. It determines the probability of collision based on the future movement trajectory and vehicle trajectory. By calculating the collision probability, it provides the driver with a more intuitive risk assessment. In actual applications, the collision probability is adjusted according to the actual situation to adapt to different driving scenarios. Finally, the warning module sends a warning signal to the driver through at least one warning channel based on the collision probability. The warning channels include sound and light, which are designed to remind the driver of potential collision risks. The issuance of the warning signal provides the driver with enough time to respond and avoid or reduce the occurrence of collision accidents.

[0144] In short, this machine vision-based car anti-collision warning system realizes real-time monitoring of the vehicle's surrounding environment and collision risk assessment through the collaborative work of multiple modules. It helps to improve driving safety, reduce the incidence of traffic accidents, and contribute to the development of my country's transportation industry. In the future, with the continuous advancement of technology, this warning system will be more perfect and bring people a safer driving experience.

[0145] Third embodiment

[0146] Based on the first embodiment, the present invention also provides a machine vision-based automobile anti-collision warning device, including a processor, an interface and a memory, wherein the processor and the interface are connected to the memory and the automobile expansion port, wherein the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute the machine vision-based automobile anti-collision warning method.

[0147] The present invention proposes a machine vision-based automobile anti-collision warning device, which has important application value in the field of modern automobile driving safety. The main components of the device include a processor, an interface and a memory. These modules cooperate with each other to jointly realize the function of automobile anti-collision warning. The processor, as the core of the device, is responsible for calling the computer program in the memory and performing data processing. These computer programs contain a series of program instructions for realizing the anti-collision warning method based on machine vision. The connection between the processor and the memory, the interface and the automobile expansion port enables the computer program to run smoothly in the device and effectively monitor various situations during the driving process of the car in real time. The function of the memory is to store computer programs and provide a basis for the processor to perform tasks. The program instructions in the memory cover a variety of warning scenarios, such as the vehicle driving too fast, the distance to the front vehicle is too close, and the oncoming vehicle. These instructions help the driver to timely discover potential collision risks and improve driving safety. The interface connects the processor, the memory and the automobile expansion port to realize bidirectional data transmission. Through the interface, the automobile anti-collision warning device obtains real-time data during the driving process of the car, such as the vehicle position, speed, and direction, and also transmits the processed warning information to the driver so that timely measures can be taken to avoid collision.

[0148] In summary, the machine vision-based automobile anti-collision warning device proposed in the present invention realizes real-time monitoring and warning of collision risks during automobile driving through the coordinated work of the processor, memory, interface and automobile expansion port. This not only helps to improve the driving safety of drivers, but also helps to alleviate the occurrence of urban traffic accidents, and has broad application prospects.

[0149] Fourth embodiment

[0150] Based on the first embodiment, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the machine vision-based automobile anti-collision warning method.

[0151] Based on the first embodiment, the present invention further provides a computer-readable storage medium. This computer-readable storage medium has the function of storing computer programs, and these computer programs contain a series of program instructions. When these instructions are executed by a processor, they will guide the processor to implement a vehicle anti-collision warning method based on machine vision.

[0152] The present invention provides a machine vision-based automobile anti-collision warning method, which realizes real-time monitoring of the vehicle's surrounding environment through computer programs and processors, provides accurate and timely warning information to the driver, and effectively improves driving safety. This technical solution has significant advantages in real-time, accuracy, adaptability, reliability and user-friendliness, and is expected to become a key technology in the field of future automobile safety assisted driving.

[0153] The above disclosure is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. A person skilled in the art will understand that all or part of the processes of the above embodiment and the changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. Automobile anti-collision warning method based on machine vision, It is characterized in that Including: using multiple high-precision cameras to obtain images of the vehicle's surrounding environment; estimating the estimated position of the moving objects around the vehicle based on the environment image; The estimated position of the moving object is corrected by using the ranging radar to obtain a corrected position; Predict the trajectory of a moving object based on the corrected position extracted from consecutive video frames; Determine the collision probability based on the future movement trajectory and vehicle trajectory; Sending a warning signal to the driver through at least one warning channel according to the collision probability; the specific steps of correcting the estimated position of the moving object by the ranging radar to obtain the corrected position include: Determine the position of the moving object detected by the radar in the radar coordinate system; Converting the position data in the radar coordinate system into coordinates corresponding to the image coordinate system captured by the vehicle camera; Compare the position data provided by the radar with the position data estimated based on the environment image; The estimated position is corrected according to the comparison result to obtain the corrected position.

2. The machine vision-based automobile anti-collision warning method according to claim 1, characterized in that: The specific steps of using multiple high-precision cameras to acquire images of the vehicle's surrounding environment include: Continuously capture surrounding environment images through multiple high-precision cameras; transmitting the captured environment image from the camera to the central processing unit; The environment image is preprocessed by the central processing unit.

3. The machine vision-based automobile anti-collision warning method according to claim 2, characterized in that: The specific step of estimating the estimated position of the moving object around the vehicle based on the environment image comprises: Apply the YOLO object detection algorithm to identify moving objects in the preprocessed environment image; Calculate the center point of the moving object and generate its position in the image; The depth information of the moving object is estimated by multi-view geometric fusion of multiple high-precision cameras, thereby obtaining the estimated position of the moving object in three-dimensional space.

4. The machine vision-based automobile anti-collision warning method according to claim 3, characterized in that: The specific steps of predicting the moving trajectory of the moving object based on the corrected position extracted from continuous video frames include: Using the position data in consecutive frames, the trajectory of the moving object is constructed; The Kalman filter is applied based on the trajectory of the moving object to predict the future movement trajectory of the moving object.

5. The machine vision-based automobile anti-collision warning method according to claim 4, characterized in that: The specific steps of applying the Kalman filter to predict the future movement trajectory of the moving object based on the moving object trajectory include: Choose two-dimensional variables to describe the position and velocity of the moving object; Setting the state transfer equation and observation model under the condition that the acceleration is constant during a preset evaluation period; Initialize the Kalman filter based on the initial observation data; Use the observation model to predict the current state based on the state estimate at the previous moment.

6. The machine vision-based automobile anti-collision warning method according to claim 5, characterized in that: The specific steps of determining the probability based on the future moving trajectory and the vehicle trajectory include: Calculate the minimum distance and time interval between the two trajectories based on the future moving trajectory and the vehicle trajectory; If a potential collision is detected, the probability of collision is assessed based on the number of possible avoidance measures; Risks are graded according to the probability of collision.

7. A vehicle anti-collision warning system based on machine vision, using the vehicle anti-collision warning method based on machine vision according to any one of claims 1 to 6, It is characterized in that It includes: image data acquisition module, position estimation module, radar correction module, movement trajectory prediction module, collision probability judgment module and early warning module; The image data acquisition module is used to acquire images of the vehicle's surrounding environment using multiple high-precision cameras; The position estimation module is used to estimate the estimated position of the moving objects around the vehicle based on the environment image; The radar correction module is used to correct the estimated position of the moving object by using the ranging radar to obtain a corrected position; The movement trajectory prediction module is used to predict the movement trajectory of the moving object based on the corrected position extracted from the continuous video frames; The collision probability judgment module is used to judge the collision probability based on the future movement trajectory and the vehicle trajectory; The warning module is used to send a warning signal to the driver through at least one warning channel according to the collision probability.

8. A vehicle anti-collision warning device based on machine vision, characterized in that: It comprises a processor, an interface and a memory, wherein the processor and the interface are connected to the memory and a vehicle expansion port, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the machine vision-based vehicle anti-collision warning method as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor executes the machine vision-based automobile anti-collision warning method as claimed in any one of claims 1 to 6.

Images