BSD anomaly monitoring method based on integrated acousto-optic alarm

Abstract

The application provides a BSD abnormality monitoring method based on an integrated sound-light alarm, and belongs to the field of automatic driving.The BSD radar or camera is respectively installed on the left and right sides of a vehicle, and a sound-light vibration alarm is integrated in the control system of the vehicle.Firstly, an abnormality recognition model of machine learning is created, then the BSD system monitors the environment around the vehicle in real time, and the abnormality recognition model analyzes the driving behavior and the environment.Once the model predicts an abnormality, the alarm sends a sound-light vibration signal, and the automatic driving system adjusts the driving strategy in time according to the early warning signal to control the vehicle.This method can improve the safety of the automatic driving vehicle and reduce the risk of traffic accidents.

Description

Technical Field

[0001] The present invention belongs to the field of autonomous driving, and more specifically relates to a BSD anomaly monitoring method based on an integrated sound and light alarm. Background Art

[0002] The Blind Spot Detection (BSD) system is now widely used in automobile safety driving assistance systems. It detects the presence of stray objects or other vehicles in the blind spots on the left and right sides of the vehicle, mainly to reduce or avoid accidents caused by blind spots.

[0003] Existing blind spot detection systems mostly use radar or cameras as detection devices. The data they collect primarily includes the distance between the vehicle and other objects, relative speed, and multi-dimensional features such as the object's size, shape, and color. This data is fed into a pre-defined anomaly detection model for training and prediction, providing early warning of potential dangers or abnormal behavior to ensure driving safety.

[0004] However, most existing systems suffer from insufficiently refined model training, low prediction accuracy, and inability to effectively detect anomalies in complex and ever-changing real-world driving environments. They often fail to meet the real-time demands of driving environments, and their single alarm method can easily lead to drivers ignoring them, failing to achieve the desired warning effect. Summary of the Invention

[0005] This invention employs a BSD anomaly monitoring method based on an integrated sound, light, and vibration alarm. It primarily uses the Isolation Forest algorithm as an anomaly detection model, specifically training the model on multi-dimensional features collected by BSD radar or cameras to improve prediction accuracy. It also optimizes the alarm method, employing a trinity of sound, light, and vibration to alert drivers, improving the timeliness and effectiveness of early warnings.

[0006] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the method comprises:

[0007] Install BSD radars or cameras on the left and right sides of the vehicle, and integrate sound, light, and vibration alarms into the vehicle's control system;

[0008] Create an anomaly recognition model. Use historical driving data to create a machine learning anomaly recognition model.

[0009] Real-time monitoring and prediction,The BSD system monitors the environment around the vehicle in real time;,while;

[0010] The autonomous driving system adjusts the driving strategy in time according to the early warning signal, and controls the vehicle according to the alarm signal.

[0011] In one scheme, the BSD radar is installed on the left and right sides of the vehicle.

[0012] The packaged photoacoustic vibration alarm is integrated in the control system of the vehicle, and the alarm is installed on the instrument panel.

[0013] In one scheme, the creation of the abnormal identification model comprises:

[0014] Collecting historical driving data collected by BSD radar or camera and other devices, including the distance between the vehicle and other objects, the relative speed, the size, shape and color of the object, and the multi-dimensional features;

[0015] Defining an Isolation Forest algorithm model, and identifying abnormal points according to the isolation property; the evaluation function of the model is represented as:

[0016] The path length h(x) is calculated by the following formula:

[0017] h(x)=E(n)+c(n)

[0018] Wherein, E(n) is the path length from the root node to the leaf node in the isolation forest, n is the amount of samples, and c(n) is the correction factor of the sample amount n, and the formula is:

[0019] c(n)=2*(ln(n-1)+0.5772156649)-(2*(n-1) / n)

[0020] In the model, this path length is used to distinguish normal data and abnormal data: samples with shorter paths are considered to be abnormal samples, and samples with longer paths are considered to be normal samples;

[0021] Model training: input the preprocessed historical driving data into the model for training;

[0022] Model evaluation: use the model to predict the training set and the validation set, and compare the prediction results with the actual results;

[0023] Model optimization: based on the results of model evaluation, adjust the parameters of the model, the number of trees, and the isolation degree of samples, and optimize the performance of the model.

[0024] In one scheme, the model training comprises:

[0025] (1) Randomly select d feature subsets to construct d isolation trees to form an isolation forest;

[0026] (2) In each isolated tree species, a feature is selected, and then a value between the minimum and maximum values of the feature is randomly selected as a split point; the data with a feature value less than the split point is divided into the left subtree of the current node, and the data with a feature value greater than or equal to the split point is divided into the right subtree;

[0027] (3) Repeat (2) until the number of samples contained in a single leaf node is less than a certain threshold or the depth of the tree reaches a set maximum value.

[0028] In one scheme, the index of model evaluation is the outlier factor, and the calculation formula is:

[0029] OF(x)=2 -E[h(x)]

[0030] Where E[h(x)] is the average path length of sample x in all isolated trees, and the closer the outlier factor is to 1, the more likely sample x is abnormal. A threshold is set to distinguish normal samples and abnormal samples.

[0031] In one scheme, the real-time monitoring and prediction includes:

[0032] The BSD system obtains the environmental information around the vehicle in real time through radar or camera, including the position, speed and driving direction data of other vehicles; these data are represented as a multi-dimensional vector D=[x1,x2,...xn], where n is the number of features, and each xi represents a feature;

[0033] Data processing, filtering noise through Kalman filter;

[0034] The preprocessed data is input into the isolated forest model that has been trained, and the model calculates an anomaly score for each sample, i.e. the outlier factor OF; after calculating the OF value of each sample, a threshold τ is set, and when the OF value is greater than τ, the sample is considered abnormal, i.e. there is a BSD abnormality;

[0035] When the model predicts an abnormal sample, an alarm is triggered; the alarm information is the sound and light vibration signal emitted by the sound and light vibration alarm.

[0036] In one scheme, the adjustment of the driving strategy includes:

[0037] When the BSD radar detects an abnormal phenomenon and issues an alarm through the sound and light vibration alarm, the autonomous driving system begins to respond;

[0038] The autonomous driving system analyzes the signal to analyze the specific factors that cause the anomaly, whether it is due to an obstacle, a pedestrian suddenly crossing the road, too frequent lane switching, or other possible driving abnormalities;

[0039] The autonomous driving system uses real-time monitoring data from BSD radar or cameras, including the vehicle's current position, speed, direction of travel, distance to other objects, and trends in these data, to obtain specific vehicle status and road environment information.

[0040] The autonomous driving system will perform emergency braking, automatic obstacle avoidance, automatic speed adjustment, and perform these control operations;

[0041] After receiving the sound, light and vibration alarm signals, the driver takes further action based on the response of the automatic driving system.

[0042] In one embodiment, the model optimization is achieved by selecting optimal parameters, wherein the main parameters are the number of trees and the maximum depth of each tree;

[0043] The number of trees is selected using cross-validation to select the optimal number of trees;

[0044] The maximum depth of each tree. For each isolated tree, its maximum depth limits the number of splits of the sample in the tree. The maximum depth is selected by cross-validation.

[0045] The objective function is as follows:

[0046] min L(n,d)=C(V(n,d))+lambda*C(n)

[0047] Where L(n,d) is the loss function, which represents the performance metric when there are n trees, each with a depth of d; V(n,d) is the performance of the model on the validation set; C(n) represents the model complexity, the number of trees, n; lambda is the regularization coefficient, which is used to balance performance and complexity; and the solution is solved using a greedy algorithm.

[0048] Beneficial effects of the present invention:

[0049] This paper proposes a BSD anomaly monitoring method based on integrated sound and light alarms. This method uses a BSD radar or camera to monitor the vehicle's surroundings in real time. It then integrates a machine-learning anomaly recognition model to analyze and predict driving behavior and the environment in real time. Furthermore, it uses sound, light, and vibration alarms to issue real-time alerts, allowing the autonomous driving system to respond immediately to BSD anomalies and adjust driving strategies. This method effectively improves the accuracy of BSD anomaly monitoring and the timeliness of early warnings, significantly reducing traffic accidents caused by driver blind spots, thereby improving driving safety and protecting the lives of drivers and passengers. It has high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Flow chart of the method of the present invention;

[0051] Figure 2 An abnormality identification model flowchart is created for the present application;

[0052] Figure 3 A real-time monitoring and prediction flowchart is created for the present application. DETAILED DESCRIPTION

[0053] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The drawings show typical embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described in the present application. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0054] Unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the present application are only for the purpose of describing the specific embodiments of the present application and are not intended to limit the present application.

[0055] As shown in Figure 1 The BSD abnormality monitoring method based on the integrated sound-light alarm includes:

[0056] S1, BSD radar or camera is installed on the left and right sides of the vehicle, and a sound-light vibration alarm is integrated in the control system of the vehicle.

[0057] First of all, BSD radar or camera needs to be installed on the left and right sides of the vehicle to detect and monitor the blind spot area of the vehicle. These two devices are generally fixed on the rearview mirror or tail light of the vehicle, and it is necessary to ensure that they have sufficient observation range and can work clearly in various weather and road conditions.

[0058] Secondly, the packaged sound-light vibration alarm should be integrated in the control system of the vehicle. The installation position of the alarm needs to fully consider the comfort of the driver and the propagation effect of the alarm signal. It is considered to be installed in a place where the driver can intuitively perceive, such as the instrument panel, so that the driver can immediately notice when the alarm signal is sent.

[0059] S2, an abnormality identification model is created, and a machine learning abnormality identification model is created through historical driving data. Isolation forest is an unsupervised learning algorithm based on decision tree, which is aimed at anomaly detection problem, and has better performance than general ensemble method. This model has higher computational efficiency when processing large data sets, and does not need to normalize the data, so it can better adapt to the data distribution of the actual scene. In addition, isolation forest is also easier to process multi-dimensional feature data sets, and can better handle anomaly value identification problems compared with distance or density based anomaly detection algorithms.

[0060] The main idea of Isolation Forest is that outliers are more easily distinguished in the decision tree structure because their positions in the feature space are often more isolated. Compared with normal data, the path length of outliers in the tree structure will be shorter, that is, outliers will be isolated earlier, which means that the identification of outliers is more sensitive. Therefore, it can be used to establish a BSD anomaly identification model to warn of possible anomalies.

[0061] As shown in Figure 2 The creation of the anomaly identification model comprises:

[0062] S201, collect historical driving data collected by BSD radar or camera and other devices, including the distance between the vehicle and other objects, relative speed, size, shape, color and other multi-dimensional features of the object. The data is preprocessed, such as denoising, missing value filling, etc., to ensure the quality of the data.

[0063] S202, define the Isolation Forest algorithm model, which is mainly based on the isolation property to identify outliers. That is, a randomly selected feature and a randomly selected cutting point value are used to recursively generate an isolated tree. The average path length of normal samples on the isolated tree will be relatively longer, while the average path length of abnormal samples will be shorter due to their distance from other samples, different characteristics, etc. Specifically, the evaluation function of the model can be expressed as

[0064] The path length h(x) is calculated by the following formula:

[0065] h(x) = E(n) + c(n)

[0066] Where E(n) is the path length from the root node to the leaf node in the isolated forest, n is the amount of samples, and c(n) is the correction factor for the sample size n, and the formula is:

[0067] c(n) = 2*(ln(n-1) + 0.5772156649) - (2*(n-1) / n)

[0068] In the model, this path length will be used to distinguish between normal data and abnormal data: samples with shorter paths are considered to be abnormal samples, and samples with longer paths are considered to be normal samples.

[0069] S203, model training: input the preprocessed historical driving data into the model for training;

[0070] The model training comprises:

[0071] (1) randomly select d feature subsets to construct d isolated trees to form an isolated forest;

[0072] The isolation forest algorithm first randomly selects a subset from all training data, and each subset is used to generate a decision tree, thereby forming a forest. Assuming that the number of selected subsets (i.e., the number of decision trees) is t, and the number of samples in the subset is ψ, then we can randomly sample ψ samples from n total samples with replacement to form t subsets. The construction of the feature subset can be achieved by the Fisher-Yates algorithm, and the time complexity of this algorithm is O(d), where d is the feature dimension.

[0073] (2) In each isolated tree, a feature is selected, and then a value between the minimum value and the maximum value of the feature is randomly selected as a split point; data with a feature value less than the split point is divided into the left subtree of the current node, and data with a feature value greater than or equal to the split point is divided into the right subtree.

[0074] Then in each isolated tree, a feature dimension i (the selection range of the feature dimension is 1-d) is randomly selected, and a split value p is randomly selected within the value range [min, max] of the feature. The decision rule is generated according to the selected feature and the corresponding split value. A feature i is randomly selected from the d features; a split value p is randomly selected between the maximum value and the minimum value of the feature i.

[0075] (3) The data is split according to the decision rule determined in step (2) to form left and right subtrees, and steps (2) and (3) are recursively performed on the left and right subtrees, respectively. Along the decision tree down, when the termination condition (the size of the data set is less than a certain threshold or the depth of the decision tree reaches the set maximum depth) is met, the splitting is stopped, and a single node is returned as a subtree.

[0076] If the size of the data set is less than a certain threshold or the depth of the decision tree reaches the maximum depth, a single node is returned as a subtree; otherwise, the data set is further divided into two parts according to the decision rule determined in step (2), and the two parts are respectively taken as the left and right subtrees of the current node.

[0077] The above is the model training process of the isolated forest. After such training, each sample has an average path length in all decision trees in the forest, and this length can be used to quantify the normality of the sample. A lower average path length means that the sample is more likely to be an outlier.

[0078] S204, model evaluation: predicting the training set and the validation set using the model, and comparing the prediction results with the actual results;

[0079] The index of the model evaluation is the outlier factor, and the calculation formula is:

[0080] OF(x)=2 -E[h(x)]

[0081] Where E[h(x)] is the average path length of sample x in all isolated trees, the closer the outlier factor is to 1, the more likely the sample x is an anomaly. By setting a threshold, normal samples and abnormal samples can be distinguished.

[0082] S205, model optimization: based on the results of model evaluation, adjust the parameters of the model, the number of trees, the degree of isolation of samples, and optimize the performance of the model. The model optimization is achieved by selecting the best parameters, and the main parameters are two, the number of trees and the maximum depth of each tree.

[0083] Number of trees: cross-validation is used to select the optimal number of trees. For example, the data set is divided into training set and validation set, and different number of trees are trained and validated, and the performance of the model (such as AUC, accuracy, etc.) is recorded in each case, and finally the number of trees with the best performance is selected. However, it should be noted that the increase of the number of trees will increase the complexity of the calculation, so in practical application, the performance and the complexity of the calculation need to be balanced.

[0084] Maximum depth of each tree: for each isolated tree, its maximum depth limits the number of sample splitting in the tree. Generally, the deeper the depth, the stronger the expression ability of the model, but too deep tree may cause overfitting. Therefore, the selection of maximum depth also needs to be carried out through cross-validation method. The specific process is similar to the selection of the number of trees, and the performance of the model under different depths is recorded, and finally the depth with the best performance is selected.

[0085] The objective function is as follows:

[0086] min L(n,d)=C(V(n,d))+lambda*C(n)

[0087] Where L(n,d) is the loss function, representing the performance measure of n trees, each tree with depth d; V(n,d) is the performance of the model on the validation set, C(n) represents the model complexity, the number of trees n; lambda is the regularization term coefficient, used to balance the performance and complexity; solved by greedy algorithm.

[0088] As shown in Figure 3 S3, real-time monitoring and prediction, the BSD system monitors the environment around the vehicle in real time.

[0089] S301, data collection: the BSD system acquires the environmental information around the vehicle in real time through radar or camera, including the position, speed, driving direction and other data of other vehicles. These data can be represented as a multi-dimensional vector D=[x1,x2,...xn], where n is the number of features, and each xi represents a feature, such as position or speed.

[0090] S302, data processing: Since the real-time collected data contains noise, preprocessing such as denoising, normalization, etc. is needed to make the data more suitable for model analysis. For example, filter the noise through a filter, such as Kalman filter.

[0091] S303, model prediction: input the preprocessed data into the isolated forest model that has been trained, the model will calculate an anomaly score (outlier factor, OF) for each sample. The OF calculation formula of isolated forest model is: OF(x) = 2^-E(h(x)), where E(h(x)) is the average path length of sample x in all isolated trees. After calculating the OF value of each sample, a threshold τ can be set, when the OF value is greater than τ, it is considered that this sample is an abnormal sample, that is, there is BSD abnormal situation.

[0092] S304, abnormal early warning: when the model predicts an abnormal sample, an alarm is triggered. The alarm information can be the sound and light vibration signal of the sound and light vibration alarm, or the abnormal information displayed by the control system.

[0093] S4, the automatic driving system adjusts the driving strategy in time according to the early warning signal, and controls the vehicle according to the alarm signal.

[0094] When the BSD system detects abnormal phenomena and sends an alarm through the sound and light vibration alarm, the automatic driving system starts to respond.

[0095] The automatic driving system will immediately receive this abnormal alarm signal and start to analyze the signal to understand the specific factors that cause the anomaly, such as obstacles, pedestrians suddenly crossing the road, too frequent lane switching or other possible driving abnormalities.

[0096] At the same time, the automatic driving system will refer to the real-time monitoring data from the BSD radar or camera, such as the current position, speed, direction of travel, distance from other objects, and the trend of these data, to obtain specific vehicle state and road environment information.

[0097] Based on this information, the automatic driving system will select the most appropriate processing method according to the preprogrammed decision strategy, such as emergency braking, automatic obstacle avoidance, automatic speed adjustment, etc., and execute these control operations, so that the vehicle can still maintain stability and safety under abnormal conditions.

[0098] After receiving the sound and light vibration alarm signal, the driver can make further operations based on the response already made by the automatic driving system, such as cooperating with the control of the automatic driving system to make emergency braking, steering to avoid, etc., so that the safety of driving is further guaranteed.

[0099] Embodiment one

[0100] The BSD anomaly of a certain automobile is monitored and alarmed by the method.

[0101] Firstly, one BSD radar is installed on each of the left and right sides of the automobile, and an audible and visual alarm is integrated into the control system of the automobile, which is installed at a position that can be clearly seen and heard by the driver, such as the instrument panel.

[0102] Next, historical driving data is collected for model training. For each drive, the driving data of the vehicle is collected by the BSD radar, including the distance between the vehicle and other objects, the relative speed, the size, shape, color, etc. of the objects. For example, the driving data records the distance, speed, etc. of 1000 other vehicles. After preprocessing, these data form a training set containing 1000 samples. Each sample is a multi-dimensional vector, and each dimension corresponds to a feature. For example, the sample (50m, 60km / h, 1.8m, 4.5m, "red") represents a distance of 50 meters from other vehicles, a relative speed of 60 km / h, and information such as the length of the other vehicle of 1.8 meters, the width of 4.5 meters, and the color of red.

[0103] Then, an isolation forest model is created for training. Here we set the parameters of the isolation forest model as follows: the number of trees is 100, and the maximum depth of each tree is 8. According to these parameters, an isolation forest model is created and input into the training set for model training.

[0104] After training, the driving behavior and environment can be monitored and predicted in real time. For example, in a certain drive, the BSD radar obtains real-time environmental information around the vehicle, such as detecting an object 30 meters away on the right side at a relative speed of 80 km / h, with a size of 2.0m*1.5m and a color of blue. This multi-dimensional vector is input into the model to obtain an anomaly score of 0.65, which exceeds the preset threshold of 0.5, so the sample is judged to be an abnormal sample, and the audible and visual alarm immediately sends a warning signal.

[0105] The automatic driving system of the automobile receives the signal, analyzes the signal, and finds that there is an object approaching rapidly on the right side. Therefore, the driving strategy is adjusted, and a decision is made to slow down and avoid based on the real-time monitoring data of the radar, thus avoiding a possible collision accident in time.

[0106] Therefore, the method well plays its function in BSD anomaly monitoring and alarming, and ensures the safety of driving.

[0107] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, the program can include the processes of the above-mentioned embodiment methods. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM), etc.

[0108] It should be understood that the above detailed description of the technical solutions of the present application by means of preferred embodiments is illustrative rather than limiting. Those skilled in the art can modify the technical solutions recorded in each embodiment or make equivalent replacement for part of the technical features on the basis of the description of the present application; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A BSD anomaly monitoring method based on an integrated sound and light alarm, characterized in that: The method includes: Install BSD radars or cameras on the left and right sides of the vehicle, and integrate sound, light, and vibration alarms into the vehicle's control system; Create an anomaly recognition model and create a machine learning anomaly recognition model; Real-time monitoring and prediction: real-time monitoring of the environment around the vehicle; The autonomous driving system adjusts the driving strategy in time according to the warning signal, and controls the vehicle according to the alarm signal; The creation of the anomaly recognition model includes: Collect historical driving data collected by BSD radar or cameras, including the distance between the vehicle and other objects, relative speed, and multi-dimensional features of the objects such as size, shape, and color; Define the Isolation Forest algorithm model to identify outliers based on isolation properties; the evaluation function of the model is expressed as: The path length h(x) is calculated by the following formula: ; Where E(n) is the path length from the root node to the leaf node in the isolation forest, n is the number of samples, and c(n) is the correction factor for the sample size n. The formula is: ; In the model, this path length is used to distinguish normal data from abnormal data: samples with shorter paths are considered abnormal samples, and samples with longer paths are considered normal samples; Model training: Input the preprocessed historical driving data into the model for training; Model evaluation: Use the model to make predictions on the training set and validation set, and compare the predicted results with the actual results; Model optimization: Based on the results of model evaluation, adjust the model parameters, the number of trees, the degree of sample isolation, and optimize the model performance; The model optimization is achieved by selecting the best parameters, which are the number of trees and the maximum depth of each tree. The number of trees is selected using cross-validation to select the optimal number of trees; The maximum depth of each isolated tree limits the number of splits of a sample in the tree; the maximum depth is selected by cross-validation; The objective function is as follows: ; Where L(n, d) is the loss function, representing the performance metric when there are n trees, each with a depth of d; V(n, d) is the performance of the model on the validation set; C(n) represents the model complexity, the number of trees, n; lambda is the regularization coefficient, used to balance performance and complexity; and the solution is obtained using a greedy algorithm.

2. The BSD anomaly monitoring method based on an integrated sound and light alarm according to claim 1, characterized in that: The BSD radar is installed on the left and right sides of the vehicle; An encapsulated sound, light and vibration alarm is integrated into the vehicle's control system; the alarm is installed on the instrument panel.

3. The BSD anomaly monitoring method based on an integrated sound and light alarm according to claim 1, characterized in that: The model training includes: (1) Randomly select d feature subsets and construct d isolation trees to form an isolation forest; (2) In each isolation tree, select a feature and then randomly select a value between the minimum and maximum values ​​of the feature as the split point; the data with a feature value less than the split point is divided into the left subtree of the current node, and the data with a feature value greater than or equal to the split point is divided into the right subtree; (3) Repeat (2) until the number of samples contained in a single leaf node is less than a certain threshold or the depth of the tree reaches the set maximum value.

4. The BSD anomaly monitoring method based on an integrated sound and light alarm according to claim 1, characterized in that: The indicator of model evaluation is the outlier factor, which is calculated as follows: ; Among them, E[h(x)] is the average path length of sample x in all isolated trees. The closer the outlier factor is to 1, the more abnormal the sample x is. A threshold is set to distinguish normal samples from abnormal samples.

5. The BSD anomaly monitoring method based on an integrated sound and light alarm according to claim 1, characterized in that: The real-time monitoring and prediction include: Real-time acquisition of environmental information around the vehicle, including the location, speed, and driving direction of other vehicles; represented as a multidimensional vector D=[x1,x2,...xn], where n is the number of features and each xi represents a feature; Data processing, filtering data noise through Kalman filter; The processed data is input into the trained isolation forest model, and an outlier score, namely the outlier factor (OF), is calculated for each sample. After calculating the OF value of each sample, a threshold τ is set. When the OF value is greater than τ, the sample is considered an outlier, that is, a BSD anomaly exists. When the model predicts an abnormal sample, an alarm is triggered; the alarm information is the sound, light and vibration signal emitted by the sound, light and vibration alarm.

6. The BSD anomaly monitoring method based on an integrated sound and light alarm according to claim 1, characterized in that: The adjusting driving strategy includes: When the BSD radar detects an abnormality and issues an alarm through the sound, light, and vibration alarm, the autonomous driving system begins to respond; The autonomous driving system interprets this signal and analyzes the specific factors that caused the abnormality, including obstacles, pedestrians suddenly crossing the road, excessive lane changes, or other driving anomalies. The autonomous driving system uses real-time monitoring data from BSD radar or cameras, including the vehicle's current position, speed, direction of travel, distance to other objects, and trends in these data, to obtain specific vehicle status and road environment information. The autonomous driving system provides emergency braking, automatic obstacle avoidance, and automatic speed adjustment; After receiving the sound, light and vibration alarm signals, the driver takes further action based on the response of the automatic driving system.

Images