Abnormal static vehicle identification method, medium, equipment and vehicle

Abstract

The invention discloses an abnormal static vehicle identification method, a medium, equipment and a vehicle. The method comprises the following steps: constructing a data set by using an abnormal static vehicle positive example meeting a first preset condition and an abnormal static vehicle negative example meeting a second preset condition; wherein the first preset condition is a first target vehicle which is located in a road lane in a preset area range in front of the automatic driving vehicle and is kept static in both historical observation duration and future observation duration; the second preset condition is a second target vehicle which is located in a road lane in a preset area range in front of the automatic driving vehicle, keeps still in historical observation duration and moves in future observation duration; training a preset dichotomy model by adopting the data set, wherein the dichotomy model is built based on a deep learning model; and deploying the trained dichotomy model into the automatic driving system to identify an abnormally static vehicle. Through the above mode, the identification accuracy and flexibility of the abnormal static vehicle can be improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving, and in particular to an abnormally stationary vehicle identification method, medium, device and vehicle. Background Art

[0002] The existing technology for identifying abnormal stationary vehicles in intelligent driving systems is mainly based on rule-based methods, relying on the perceived and observed object motion history and semantic map information to identify abnormal stationary vehicles.

[0003] The rule-based abnormal stationary vehicle identification method is limited by the prior knowledge of traffic regulations and other aspects provided by technical personnel, and has limitations in design and performance. For complex traffic scenarios, special events such as temporary traffic control are not covered by traditional rules, resulting in the inability to flexibly identify and adapt well to various road conditions, lacking flexibility. It cannot be widely used in areas with changing traffic conditions, such as areas with different traffic regulations, and has poor versatility. In atypical situations, such as vehicles parked in non-specified areas due to emergencies or special reasons, the rule-based system may not be able to accurately identify these situations, resulting in false alarms or missed alarms. Moreover, the rule-based method may not cover all possible illegal parking situations, especially on multi-lane, multi-directional roads, or in areas with special traffic signs and signals. It cannot cover the identification of abnormal stationary vehicles in all scenarios. Summary of the invention

[0004] The present application mainly provides a method, medium, device and vehicle for identifying abnormal stationary vehicles to solve the problems of existing identification of abnormal stationary vehicles, such as reliance on rules, insufficient scene coverage, insufficient versatility, lack of flexibility and accuracy.

[0005] In order to solve the above technical problems, a technical solution adopted in the present application is: providing a method for identifying abnormal stationary vehicles based on a deep learning model, including: constructing a data set with positive examples of abnormal stationary vehicles that meet a first preset condition and negative examples of abnormal stationary vehicles that meet a second preset condition; wherein the first preset condition is a first target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during both the historical observation time and the future observation time; the second preset condition is a second target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during the historical observation time and moves during the future observation time; using the data set to train a preset binary classification model, the binary classification model is built based on a deep learning model; and deploying the trained binary classification model into the autonomous driving system to identify abnormal stationary vehicles.

[0006] The recognition accuracy of the binary classification model is improved through high-precision annotated data that meets the preset conditions. By training the preset binary classification model and deploying the trained binary classification model on the autonomous driving vehicle, the autonomous driving vehicle can efficiently identify abnormal stationary vehicles on the road, effectively improving the recognition accuracy of abnormal stationary vehicles and avoiding misjudgment and missed judgment.

[0007] In some embodiments, the training data set is constructed with positive examples of abnormal stationary vehicles that meet a first preset condition and negative examples of abnormal stationary vehicles that meet a second preset condition, including: traversing each vehicle perceived by the autonomous driving vehicle in a traffic scene; and based on the first preset condition and the second preset condition, obtaining the positive examples of the abnormal stationary vehicles and the negative examples of the abnormal stationary vehicles through data labeling and data mining.

[0008] By marking the positive and negative examples of vehicles in traffic scenes, we can accurately identify the first target vehicles that are in abnormal stationary states such as illegal parking, emergency parking or accident vehicles, and identify the second target vehicles that are in non-abnormal stationary states such as traffic jams, waiting for traffic lights or queuing. This enables the binary classification model to accurately determine the similarity between each vehicle on the road and the positive or negative examples of abnormal stationary vehicles, thereby accurately calculating the probability that each vehicle on the road is an abnormal stationary vehicle.

[0009] In some embodiments, the positive example of the abnormal stationary vehicle and the negative example of the abnormal stationary vehicle are obtained by data labeling and data mining, including: accepting that the first target vehicle is manually labeled as a positive example label, and the second target vehicle is manually labeled as a negative example label; the first target vehicle is automatically labeled as a positive example label, and the second target vehicle is automatically labeled as a negative example label; wherein the range of the preset area in front of the autonomous driving vehicle corresponding to the manual labeling is larger than the range of the preset area in front of the autonomous driving vehicle corresponding to the automatic labeling.

[0010] The first and second target vehicles are accurately labeled through manual labeling for the binary classification model to identify and learn.

[0011] In some embodiments, during the automatic labeling process, the first preset condition and the second preset condition also include that the angle formed by the orientation of the target vehicle and the orientation of the autonomous driving vehicle is within a preset acute angle.

[0012] By setting the angle formed by the direction of the target vehicle and the direction of the autonomous driving vehicle to be within a preset acute angle in the second preset condition, it is possible to avoid that irregularly parked vehicles such as vehicles parked on the sidewalk are judged as negative examples of abnormally stationary vehicles, thereby improving the training effect of the binary classification model.

[0013] In some embodiments, the positive example of the abnormal stationary vehicle includes a first target vehicle and a positive example label, and the negative example of the abnormal stationary vehicle includes a second target vehicle and a negative example label; the using the data set to train the preset binary classification model includes: obtaining a feature vector of the target vehicle; the target vehicle includes the first target vehicle and / or the second target vehicle in the data set; the feature vector is the interaction information between the autonomous driving vehicle and the target vehicle; the feature vector of the target vehicle and the motion data within the historical observation period are input into the binary classification model to obtain the probability that the target vehicle is an abnormal stationary vehicle; the label corresponding to the target vehicle is obtained, and the error between the probability and the label is calculated, wherein the label is a positive example label or a negative example label corresponding to the target vehicle; the binary classification model is corrected based on the error until the error between the probability and the label is less than a preset error threshold.

[0014] By building and training a binary classification model to identify and classify abnormal stationary vehicles on the road, autonomous vehicles can be helped to more accurately and quickly identify abnormal stationary vehicles on the road and respond accordingly. At the same time, the deep learning-based model can learn based on new data and adapt to different road conditions, making the identification of abnormal stationary vehicles more flexible.

[0015] In some embodiments, the binary classification model includes a neural convolutional network and a multi-layer perceptron network; the step of inputting the virtual lane data and the target vehicle historical data into the binary classification model to obtain the probability that the target vehicle is an abnormally stationary vehicle includes: inputting the virtual lane data and the target vehicle historical data into the neural convolutional network for feature extraction to obtain implicit features in the virtual lane data and the target vehicle historical data; inputting the implicit features into the multi-layer perceptron network for mapping transformation to obtain the probability that the target vehicle is an abnormally stationary vehicle.

[0016] By extracting data features through convolutional neural networks, the efficiency and accuracy of data feature extraction are improved, which can be applied to diverse data and has high generalization. Through the efficient computing characteristics of the multi-layer perceptron network, the mapping transformation process can be completed efficiently, meeting the real-time computing needs of road vehicle identification.

[0017] In some embodiments, deploying the trained binary classification model into the autonomous driving system includes: serializing the trained binary classification model into binary model data through an inference engine; and deploying the binary model data to the autonomous driving system through a programming language.

[0018] By serializing the binary classification model into binary model data and then deploying it, it can directly correspond to machine code. When parsing, there is no need for complex syntax and format parsing like text format, which helps to greatly improve the parsing speed and speed up the model's reasoning speed.

[0019] In order to solve the above technical problems, another technical solution adopted by the present application is: providing a storage medium on which program data is stored, and when the program data is executed by a processor, the steps of the abnormal stationary vehicle identification method as described above are implemented.

[0020] The beneficial effects of the storage medium refer to the beneficial effects of the abnormal stationary vehicle identification method mentioned above, which will not be repeated here.

[0021] The present application also provides a vehicle-mounted device, including a processor and a memory connected to each other, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the abnormal stationary vehicle identification method as described above are implemented.

[0022] The vehicle-mounted device is a computer-type device. The beneficial effects of the vehicle-mounted device refer to the beneficial effects of the above-mentioned abnormally stationary vehicle identification method, which will not be repeated here.

[0023] The present application also provides a vehicle, comprising the above-mentioned vehicle-mounted device. During the driving process, the vehicle can complete the steps of the above-mentioned abnormal stationary vehicle identification method through the control of the vehicle-mounted device.

[0024] The beneficial effects of the present application are as follows: Different from the prior art, the present application discloses an abnormal stationary vehicle identification method, medium, device and vehicle, and constructs a data set by using positive examples of abnormal stationary vehicles that meet a first preset condition and negative examples of abnormal stationary vehicles that meet a second preset condition as training reference data for a binary classification model; wherein the first preset condition is a first target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during both a historical observation period and a future observation period, and the second preset condition is a second target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during the historical observation period and moves during the future observation period, and by dividing the first and second target vehicles, the binary classification model can be classified according to the preset conditions. The software classifies the types of target vehicles on the road, and calculates the gap between them and positive examples or negative examples for learning, so that the recognition process no longer relies on prior knowledge such as traffic regulations provided by technical personnel, and has better design and higher efficiency; the data set is used to train a preset binary classification model, and the binary classification model is built based on a deep learning model, which can be widely used in areas with changing traffic conditions and has high versatility; the trained binary classification model is deployed in an autonomous driving system to identify abnormally stationary vehicles, so that the autonomous driving vehicle can effectively identify abnormally stationary vehicles on the road and respond in time, which improves the accuracy of abnormally stationary vehicle recognition, and determines the vehicle's next driving action as stopping, following or dodging, etc., to avoid collisions and improve driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0026] Figure 1 It is a flow chart of an embodiment of an abnormal stationary vehicle identification method provided by the present application;

[0027] Figure 2 Yes Figure 1 The flowchart of step 10 of an abnormal stationary vehicle identification method shown in FIG.

[0028] Figure 3 Yes Figure 2 The flowchart of step 12 of an abnormal stationary vehicle identification method is shown as an embodiment;

[0029] Figure 4 Yes Figure 2The schematic diagram of another embodiment of step 12 of the abnormal stationary vehicle identification method shown;

[0030] Figure 5 Yes Figure 1 The flowchart of step 20 of an abnormal stationary vehicle identification method according to an embodiment is shown;

[0031] Figure 6 Yes Figure 5 The flowchart of step 22 of an abnormal stationary vehicle identification method according to an embodiment is shown;

[0032] Figure 7 Yes Figure 1 The flowchart of step 30 of an abnormal stationary vehicle identification method is shown as an embodiment;

[0033] Figure 8 It is a structural schematic diagram of an embodiment of a storage medium provided by the present application;

[0034] Fig. 9 It is a structural schematic diagram of an embodiment of the vehicle-mounted device provided by the present application;

[0035] Fig.10 It is a structural schematic diagram of a vehicle embodiment provided by the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0037] The terms "first", "second", "third" in the embodiments of the present application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first", "second", "third" can expressly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0038] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0039] See also Figure 1 , Figure 1 1 is a flow chart of an embodiment of an abnormal stationary vehicle identification method provided by the present application, and the abnormal stationary vehicle identification method includes:

[0040] Step 10: Construct a data set based on positive examples of abnormal stationary vehicles that meet the first preset condition and negative examples of abnormal stationary vehicles that meet the second preset condition.

[0041] According to the vehicles in the perceived traffic scene, a first preset condition for identifying abnormal stationary vehicles and a second preset condition for identifying non-abnormal stationary vehicles are set, and abnormal stationary vehicles that meet the first preset condition are marked as positive examples, and abnormal stationary vehicles that meet the second preset condition are marked as negative examples, and a data set is constructed with the positive vehicle data and the negative vehicle data.

[0042] Among them, the first preset condition is the first target vehicle that is located in the road lane within the preset area in front of the autonomous driving vehicle and remains stationary during the historical observation time and the future observation time; the first preset condition is used to identify abnormally stationary vehicles on the road, such as vehicles parked in prohibited parking areas, vehicles illegally parked during peak hours, vehicles legally parked in parking spaces, vehicles parked in temporary parking places such as in front of overtime doors, or vehicles stopped in the driving lane for no reason, etc.

[0043] In the abnormal stationary vehicle binary classification model, positive examples usually refer to stationary vehicles that are detected as abnormal. These vehicles may be in an abnormal stationary state due to some reasons such as failure, accident, illegal parking, etc., and are the objects that the model needs to focus on identifying and paying attention to.

[0044] The second preset condition is a second target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during the historical observation period and moves during the future observation period; the second preset condition is used to identify non-abnormally stationary vehicles on the road, such as vehicles that have stopped due to congestion or waiting for traffic lights.

[0045] Corresponding to the positive examples, the negative examples in the abnormal stationary vehicle binary classification model represent those vehicles that are detected as normal or not in the category of concern. These vehicles may be driving normally, parked legally, or in other states that do not require special attention. The negative example of marking a non-abnormal stationary vehicle that meets the second preset condition as an abnormal stationary vehicle indicates that such vehicles are not considered as abnormal stationary vehicles.

[0046] Autonomous driving vehicles need to identify which vehicles are abnormally stationary and perform evasive maneuvers in a timely manner.

[0047] The vehicle samples that meet the first preset condition are taken as positive examples, and the vehicle samples that meet the second preset condition are taken as negative examples. A data set for identifying abnormally stationary vehicles is constructed, and an effective model is trained for identifying abnormally stationary vehicles, which is conducive to improving the recognition accuracy of abnormally stationary vehicles by autonomous driving vehicles.

[0048] Optionally, adjust the ratio of positive samples to negative samples in the dataset to ensure the balance of the dataset, that is, the number of positive and negative examples is similar, so as to avoid the model trained by the dataset being biased towards one category.

[0049] Further, see Figure 2 , constructing a training data set with positive examples of abnormal stationary vehicles that meet the first preset condition and negative examples of abnormal stationary vehicles that meet the second preset condition, including:

[0050] Step 11: Traverse every vehicle perceived by the autonomous vehicle in the traffic scene.

[0051] Autonomous vehicles collect data about the surrounding environment perceived in traffic scenes through various sensors installed on them, such as lidar, radar, and cameras. They use algorithms such as computer vision to analyze the processed data to detect and identify targets such as vehicles, pedestrians, and obstacles. For vehicles, it is usually necessary to identify their type, such as cars, trucks, motorcycles, etc., as well as their size, direction, and location. The autonomous driving system maintains one or more lists or data structures to store all currently perceived vehicle information.

[0052] By traversing these data structures, each vehicle can be accessed and processed in turn. For example, the distance, angle, relative speed, etc. between each vehicle and the autonomous vehicle can be calculated, and obstacle avoidance strategies or path planning can be formulated accordingly.

[0053] Step 12: Based on the first preset condition and the second preset condition, obtain positive examples of abnormal stationary vehicles and negative examples of abnormal stationary vehicles through data annotation and data mining.

[0054] Positive vehicles and negative vehicles in traffic scenes are identified according to the first preset condition and the second preset condition, and the positive and negative labels of each vehicle are accurately labeled through data labeling. At the same time, a large number of positive labels that meet the first preset condition and negative labels that meet the second preset condition are labeled through data mining, so as to obtain accurate and large-scale positive and negative vehicle examples in traffic scenes, thereby improving the quality of data sets containing positive and negative examples of abnormal stationary vehicles.

[0055] Data labeling is the process of manually labeling qualified vehicles in the database to obtain accurate label data.

[0056] Data mining is to label qualified vehicles in the database through manual or automatic labeling, and quantitatively supplement the label data obtained by data annotation.

[0057] Further, see Figure 3 , obtaining the abnormal stationary vehicle positive example and the abnormal stationary vehicle negative example by means of data annotation and data mining, including:

[0058] Step 121: accepting that the first target vehicle is manually labeled as a positive example label, and the second target vehicle is manually labeled as a negative example label.

[0059] Image processing personnel label vehicles in traffic scenes that meet the first preset conditions with positive labels, and define such vehicles as first target vehicles, i.e., abnormal stationary vehicles; image processing personnel also label vehicles in traffic scenes that meet the second preset conditions with negative labels, and define such vehicles as second target vehicles, i.e., non-abnormal stationary vehicles.

[0060] By obtaining the positive example labels manually annotated for the first target vehicle and the negative example labels manually annotated for the second target vehicle, accurately annotated vehicle label data can be obtained, providing accurate label data for subsequent model training and improving the prediction accuracy of the model.

[0061] Step 122: The first target vehicle is automatically labeled as a positive example label, and the second target vehicle is automatically labeled as a negative example label.

[0062] The automatic labeling program labels the vehicles in the traffic scene that meet the first preset condition with positive labels, and defines such vehicles as the first target vehicles, that is, abnormal stationary vehicles; the automatic labeling program also labels the vehicles in the traffic scene that meet the second preset condition with negative labels, and defines such vehicles as the second target vehicles, that is, non-abnormal stationary vehicles.

[0063] By using the positive example labels annotated by the automatic labeling program for the first target vehicle and the negative example labels annotated by the automatic labeling program for the second target vehicle, a large amount of annotated vehicle label data can be obtained, which provides sufficient label data samples for subsequent model training and is conducive to improving the prediction accuracy of the model.

[0064] Optionally, see Figure 4 The first preset condition and the second preset condition also include that the angle formed by the direction of the target vehicle and the direction of the autonomous driving vehicle is within a preset acute angle.

[0065] Taking the direction facing the front of the vehicle or the direction of vehicle travel as the orientation of the current vehicle, when the angle a formed by the orientation of the target vehicle and the orientation of the autonomous driving vehicle is within the preset acute angle, the vehicle is a road vehicle participating in traffic and needs to be considered when identifying abnormally stationary vehicles; if the angle a formed by the orientation of the target vehicle and the orientation of the autonomous driving vehicle is greater than the preset acute angle, the vehicle may be a vehicle parked on the sidewalk and does not participate in traffic, so it is not considered when identifying abnormally stationary vehicles.

[0066] Specifically, the angle a formed by the direction of the target vehicle and the direction of the autonomous driving vehicle can be 45°, 50°, 55°, 60°, 63° or 68°, etc.

[0067] By limiting the angle a formed by the orientation of the identified target vehicle and the orientation of the autonomous driving vehicle to within a preset acute angle, the interference of sidewalk parked vehicles in the identification of abnormally stationary vehicles can be eliminated, which helps to improve the model's learning efficiency for the data set.

[0068] Among them, the range of the preset area in front of the autonomous driving vehicle corresponding to the manual marking is larger than the range of the preset area in front of the autonomous driving vehicle corresponding to the automatic marking.

[0069] Manual labeling can accurately mark positive and negative examples of abnormal stationary vehicles in traffic scenes, so it can identify vehicles in a large range and ensure that each target vehicle is accurately labeled as a positive or negative example. Automatic labeling relies on deep learning or computer vision, and the recognition accuracy of target vehicles is not as good as manual labeling, so it is necessary to narrow the labeling range to ensure the accuracy of labeling.

[0070] Step 20: Use the data set to train the preset binary classification model.

[0071] The data set is used as input data and verification data to train the preset binary classification model to improve the recognition accuracy of the binary classification model for abnormal stationary vehicles.

[0072] Use the deep learning framework to build a binary classification model, including designing the architecture of the binary classification model such as the input layer, hidden layer, and output layer, and specifying the loss function and evaluation indicators to improve the performance of the binary classification model. The binary classification model is built based on the deep learning model and is used to divide the input data samples into one of two categories.

[0073] Specifically, the types of binary classification models may include: logistic regression, which maps the output of a linear model to the interval (0, 1) through a function to obtain the probability of belonging to a certain category; decision trees, which classify data into one of two categories by learning a series of rules; neural networks, which can learn complex nonlinear relationships by stacking multiple layers, and use functions in the output layer to obtain the probability of belonging to a certain category, etc.

[0074] By training the binary classification model built based on the deep learning framework through the data set, the binary classification model can learn accurate and detailed recognition methods, thereby improving the performance of the binary classification model. At the same time, a large amount of training data can reduce the model's excessive dependence on training data, reduce the risk of overfitting, and be able to adapt to new data, thereby enhancing the generalization ability of the binary classification model.

[0075] Specifically, the abnormal stationary vehicle positive example includes a first target vehicle and a positive example label, and the abnormal stationary vehicle negative example includes a second target vehicle and a negative example label.

[0076] Further, see Figure 5 , use the data set to train the preset binary classification model, including:

[0077] Step 21: Obtain the feature vector of the target vehicle.

[0078] Extract the target vehicle feature vector without label information from the dataset.

[0079] Specifically, the target vehicle includes the first target vehicle and / or the second target vehicle in the data set. The binary classification model takes the first target vehicle and / or the second target vehicle in the data set as traffic participating vehicles in the recognition range as the target vehicle, and learns and recognizes abnormal stationary vehicles for the target vehicle.

[0080] The feature vector is the interaction information between the autonomous driving vehicle and the target vehicle, such as the driving speed, relative speed, acceleration, direction, relative position between the autonomous driving vehicle and the target vehicle, the current signal light status, traffic sign recognition results, and possible driver behavior predictions.

[0081] Acquiring the feature vector of the target vehicle includes acquiring road rutting information, map navigation information, relative position characteristics between the autonomous driving vehicle and the target vehicle, motion state characteristics, environmental perception characteristics, and behavior prediction characteristics, and combining the extracted features into a feature vector.

[0082] Extracting feature vectors can eliminate redundant and irrelevant information for the labels to be trained, and further improve the learning and processing efficiency of the binary classification model for data.

[0083] Step 22: Input the feature vector of the target vehicle and the motion data within the historical observation period into the binary classification model to obtain the probability that the target vehicle is an abnormal stationary vehicle.

[0084] The feature vector of the target vehicle and the speed, acceleration, position change and direction obtained from the motion data within the historical observation period are input into the binary classification model. The binary classification model calculates the displacement of the vehicle during the historical observation period, or converts the position information into an offset relative to a reference point such as the center line of the road, an intersection, etc., calculates the rate of change or standard deviation of the speed to reflect the stability of the speed, or predicts the driving trajectory of the target vehicle by extracting the rate of change of the acceleration value direction, so as to determine whether it will change from stationary to moving within a fixed time window, so as to obtain the probability that the target vehicle is an abnormal stationary vehicle, thereby distinguishing abnormal stationary vehicles from non-abnormal stationary vehicles.

[0085] Specifically, the binary classification model includes neural convolutional networks and multi-layer perceptron networks.

[0086] Through the strategy of local connection and weight sharing, convolutional neural network can effectively reduce the number of network parameters, reduce computational complexity, and extract local features of target vehicles and roads in the image. Through multi-level convolution and pooling operations, convolutional neural network can extract high-level features in the image layer by layer and enhance the generalization ability of the binary classification model.

[0087] The multi-layer perceptron network is able to learn complex nonlinear relationships. Its flexibility enables it to adapt to various feature inputs and improve the classification accuracy of the binary classification model by learning nonlinear relationships in the hidden layer.

[0088] Further, see Figure 6 , the feature vector of the target vehicle and the motion data within the historical observation time are input into the binary classification model to obtain the probability that the target vehicle is an abnormal stationary vehicle, including:

[0089] Step 221: Input the feature vector of the target vehicle and the motion data within the historical observation period into the neural convolution network for feature extraction, and obtain the feature vector of the target vehicle and the implicit features in the motion data within the historical observation period.

[0090] The feature vector of the target vehicle and its motion data within the historical observation period are input into the neural convolutional network to extract features such as vehicle type and color. The feature vector of the target vehicle and the implicit features in its motion data within the historical observation period are used to improve the attention of the binary classification model to the target vehicle within the recognition range, thereby saving the computational cost of the model and improving the data processing efficiency of the binary classification model.

[0091] The neural convolutional network automatically learns and extracts local features from the data set to predict the short-term motion pattern or trend of the target vehicle for subsequent analysis, reducing the complexity and subjectivity of feature engineering and improving the generalization ability and accuracy of the model.

[0092] Step 222: Input the implicit features into the multi-layer perceptron network for mapping transformation to obtain the probability that the target vehicle is an abnormal stationary vehicle.

[0093] The implicit features output by the neural convolutional network are mapped to the interval (0, 1) using activation functions such as sigmoid, generating a floating point number between 0 and 1, which represents the probability that the target vehicle is an abnormally stationary vehicle.

[0094] Step 23: Obtain a label corresponding to the target vehicle and calculate the error between the probability and the label, where the label is a positive label or a negative label corresponding to the target vehicle.

[0095] Get the positive or negative label of the data set corresponding to the target vehicle with the current calculated probability value, calculate the error between the probability and the label, and correct the binary classification model according to the error.

[0096] For example, the output of the binary classification model for vehicle A is 0.78, which means that the probability that vehicle A is an abnormal stationary vehicle is 0.78. In the data set, the positive label is assigned a value of 1 and the negative label is assigned a value of 0. Vehicle A is marked as a positive label in the data set, so the difference between 0.78 and 1 is calculated and returned to update the data of the binary classification model.

[0097] Step 24: Modify the binary classification model based on the error until the error between the probability and the label is less than a preset error threshold.

[0098] The error threshold is defined as the standard for evaluating the prediction ability of the binary classification model. When the error between the probability and the label is less than the preset error threshold, the binary classification model meets the training requirements. Based on the error, the parameters of the binary classification model are modified until the error between the probability and the label is less than the preset error threshold.

[0099] For example, the difference between the output of the binary classification model corresponding to vehicle A is 0.78 and the corresponding positive example label value 1 is 0.22. Based on the difference of 0.22, the parameters of the binary classification model are updated until the output of the binary classification model corresponding to vehicle A is 0.98, which is less than the preset threshold value of 0.03. At this time, the binary classification model meets the training requirements.

[0100] By setting a preset threshold, the binary classification model can not only accurately fit the training data, but also generalize to new data, which is beneficial to improving the performance of the binary classification model.

[0101] Step 30: Deploy the trained binary classification model into the autonomous driving system to identify abnormally stationary vehicles.

[0102] A binary classification model is deployed in the autonomous driving system, and road information such as image type information or sensor type information is obtained through the autonomous driving vehicle. The binary classification model uses the obtained road information to perform abnormal stationary analysis on target vehicles within the recognition range on the current road to determine the probability that each target vehicle is an abnormally stationary vehicle.

[0103] Specifically, the road information includes the driving speed, relative speed, acceleration, direction, relative position between the autonomous driving vehicle and the target vehicle, the current traffic light status, traffic sign recognition results, and possible driver behavior predictions.

[0104] By successfully deploying the trained binary classification model into the autonomous driving system to identify abnormally stationary vehicles, autonomous driving operations such as dodging or stopping can be performed, thereby improving the safety and reliability of the autonomous driving system.

[0105] Further, see Figure 7 , deploy the trained binary classification model into the autonomous driving system, including:

[0106] Step 31: Serialize the trained binary classification model into binary model data through the inference engine.

[0107] The binary model is exported from the training framework of the binary classification model, and the trained binary classification model is converted into a format more suitable for deployment in the autonomous driving system through the inference engine, forming binary model data with smaller size, faster loading speed and higher operating efficiency.

[0108] The inference engine is responsible for executing the model's inference task, that is, processing the real-time road information in the autonomous driving system through the loaded binary classification model and outputting the corresponding classification results. The inference engine is a bridge between the binary classification model and the actual application scenario, which enables the trained binary classification model to perform efficient inference on different hardware and software platforms.

[0109] Binary model data is binary classification model data saved in binary format, which is more compact and efficient than text or source code format. It can be directly read by machines and only stores the weights, structure, and other necessary information of the model without including any human-readable tags or annotations.

[0110] Serializing the binary classification model into binary model data removes all unnecessary formatting and markup, saving resource space on the device where the autonomous driving system is located. Binary model data can be directly processed by the computer's memory and processor, so loading binary model data is usually much faster than parsing text or source code formats, helping to reduce latency during model inference. Binary model data can maintain a consistent format on different operating systems and hardware architectures, and can be easily deployed and run in a variety of environments without additional conversion or modification, making it more cross-platform compatible.

[0111] Step 32: Deploy the binary model data to the autonomous driving system via a programming language.

[0112] After using the C++ API provided by the inference engine to load the binary model file, the road information data collected by the autonomous driving system, such as camera images, is preprocessed into the format expected by the binary model. The preprocessed data is then passed to the inference engine to perform inference and obtain the output results, completing the deployment of the binary model data on the autonomous driving system.

[0113] Autonomous driving vehicles can identify abnormal stationary vehicles on the road through binary model data in the autonomous driving system and respond in a timely manner to ensure the safety of autonomous driving vehicles in traffic environments.

[0114] See also Figure 8 , Figure 8 It is a structural diagram of an embodiment of the storage medium provided by the present application.

[0115] The storage medium 300 stores program data 301. When the program data 301 is executed by the processor, the following is achieved: Figure 1 The steps of the abnormal stationary vehicle identification method are described.

[0116] The program data 301 is stored in a storage medium 300, and includes a number of instructions for enabling a network device (such as a router, a personal computer, a server, or other network device) or a processor to execute all or part of the steps of the method described in each embodiment of the present application.

[0117] Optionally, the storage medium 300 may be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other medium that can store program data.

[0118] See also Fig. 9 , Fig. 9 It is a structural schematic diagram of an embodiment of the vehicle-mounted device provided by the present application.

[0119] The vehicle-mounted device 400 is installed in the vehicle and is a computer-type device. It works in conjunction with various sensor modules and control modules in the vehicle to achieve visual perception, sensor data processing, and image display. The vehicle-mounted device 400 includes a processor 420 and a memory 410 that are interconnected. The memory 410 stores a computer program. When the processor 420 executes the computer program, the steps of the above-mentioned abnormally stationary vehicle identification method are implemented.

[0120] See also Fig.10 , Fig.10 A vehicle 500 is provided, comprising the vehicle-mounted device 400 as described above. The vehicle 500 can complete the process of abnormal stationary vehicle identification through the control of the vehicle-mounted device 400 during driving.

[0121] Different from the prior art, the present application provides an abnormal stationary vehicle identification method, medium, device and vehicle, which labels the positive labels of abnormal stationary vehicles and the negative labels of abnormal stationary vehicles for vehicles in traffic scenes according to preset conditions, and uses the labeled data and the vehicle data of the traffic scene for the training of a binary classification model, so that the autonomous driving vehicle can efficiently identify abnormal stationary vehicles on the road, effectively improving the recognition accuracy of abnormal stationary vehicles and avoiding misjudgment and missed judgment; the binary classification model based on deep learning can learn according to new data and adapt to different road conditions, making the identification of abnormal stationary vehicles more flexible. The binary classification model accurately calculates the probability that the output target vehicle is an abnormal stationary vehicle. The autonomous driving vehicle performs autonomous driving operations according to the calculation results of the binary classification model to ensure the safety of the autonomous driving vehicle and the people in the vehicle.

[0122] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the storage medium embodiment and the computer device embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0123] The present application can be used in many general or special vehicle-mounted computing system environments or configurations, such as personal computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, network PCs, minicomputers, and distributed computing environments including any of the above systems or devices.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not performed.

[0125] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0126] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0127] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for identifying abnormal stationary vehicles based on a deep learning model, characterized in that: include: A data set is constructed with positive examples of abnormal stationary vehicles that meet a first preset condition and negative examples of abnormal stationary vehicles that meet a second preset condition; wherein the first preset condition is a first target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during both the historical observation time and the future observation time; the second preset condition is a second target vehicle that is located in a road lane within a preset area in front of the autonomous driving vehicle and remains stationary during the historical observation time and moves during the future observation time; The data set is used to train a preset binary classification model, where the binary classification model is built based on a deep learning model; The trained binary classification model is deployed in an autonomous driving system to identify abnormally stationary vehicles.

2. The abnormal stationary vehicle identification method according to claim 1, characterized in that: The constructing of a training data set using positive examples of abnormal stationary vehicles that meet the first preset condition and negative examples of abnormal stationary vehicles that meet the second preset condition includes: Traversing each vehicle perceived by the autonomous driving vehicle in the traffic scene; Based on the first preset condition and the second preset condition, the positive example of the abnormal stationary vehicle and the negative example of the abnormal stationary vehicle are obtained through data labeling and data mining.

3. The abnormal stationary vehicle identification method according to claim 2, characterized in that: The obtaining of the abnormal stationary vehicle positive example and the abnormal stationary vehicle negative example by means of data annotation and data mining includes: Accepting that the first target vehicle is manually labeled as a positive example label, and the second target vehicle is manually labeled as a negative example label; The first target vehicle is automatically labeled as a positive example label, and the second target vehicle is automatically labeled as a negative example label; Among them, the range of the preset area in front of the autonomous driving vehicle corresponding to the manual marking is larger than the range of the preset area in front of the autonomous driving vehicle corresponding to the automatic marking.

4. The abnormal stationary vehicle identification method according to claim 3, characterized in that: During the automatic labeling process, the first preset condition and the second preset condition also include that the angle formed by the orientation of the target vehicle and the orientation of the autonomous driving vehicle is within a preset acute angle.

5. The abnormal stationary vehicle identification method according to claim 1, characterized in that: The abnormal stationary vehicle positive example includes a first target vehicle and a positive example label, and the abnormal stationary vehicle negative example includes a second target vehicle and a negative example label; The adopting the data set to train a preset binary classification model includes: Acquire a feature vector of a target vehicle; the target vehicle includes the first target vehicle and / or the second target vehicle in the data set; the feature vector is interaction information between the autonomous driving vehicle and the target vehicle; Inputting the feature vector of the target vehicle and the motion data within the historical observation time into the binary classification model to obtain the probability that the target vehicle is an abnormally stationary vehicle; Obtaining a label corresponding to the target vehicle, and calculating an error between the probability and the label, wherein the label is a positive label or a negative label corresponding to the target vehicle; The binary classification model is modified based on the error until the error between the probability and the label is less than a preset error threshold.

6. The abnormal stationary vehicle identification method according to claim 5, characterized in that: The binary classification model includes a neural convolutional network and a multi-layer perceptron network; Inputting the feature vector of the target vehicle and the motion data within the historical observation time into the binary classification model to obtain the probability that the target vehicle is an abnormally stationary vehicle includes: Inputting the feature vector of the target vehicle and the motion data within the historical observation time into the neural convolution network for feature extraction, and obtaining the feature vector of the target vehicle and the implicit features in the motion data within the historical observation time; The implicit features are input into the multi-layer perceptron network for mapping transformation to obtain the probability that the target vehicle is an abnormally stationary vehicle.

7. The abnormal stationary vehicle identification method according to claim 1, characterized in that: The step of deploying the trained binary classification model into the autonomous driving system includes: Serializing the trained binary classification model into binary model data through an inference engine; Deploy the binary model data to the autonomous driving system via a programming language.

8. A storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the steps of the abnormal stationary vehicle identification method according to any one of claims 1 to 7 are implemented.

9. A vehicle-mounted device, characterized in that: The method comprises a processor and a memory connected to each other, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the abnormal stationary vehicle identification method as described in any one of claims 1 to 7 are implemented.

10. A vehicle, characterized in that: Comprising the vehicle-mounted device as claimed in claim 9.

Images