Vehicle bottom abnormity real-time detection method and device

Abstract

The invention discloses a vehicle bottom abnormity real-time detection method and device, and belongs to the technical field of vehicle detection. The vehicle bottom abnormity real-time detection method comprises the following steps: acquiring actual speeds corresponding to a target vehicle at a plurality of acquisition moments and initial vehicle bottom images corresponding to the target vehicle at the plurality of acquisition moments; processing the initial vehicle bottom image corresponding to the actual speed based on each actual speed, and acquiring a target vehicle bottom image corresponding to the target vehicle in a target acquisition time period; and processing the target vehicle bottom image to obtain a vehicle bottom detection result corresponding to the target vehicle. According to the vehicle bottom abnormity real-time detection method, distorted images caused by vehicle speed and other factors can be effectively corrected, vehicle bottom images without obvious differences can be collected, abnormal conditions existing in the vehicle bottom can be recognized based on the target vehicle bottom image, the detection efficiency is improved, and the detection precision and accuracy are improved.

Description

Technical Field

[0001] The present application belongs to the field of vehicle detection technology, and in particular, relates to a method and device for real-time detection of vehicle bottom abnormalities. Background Art

[0002] In places such as airports, customs, or major sports venues, vehicles entering and leaving need to be inspected under the vehicle to prevent prohibited items and other foreign objects from being hidden under the vehicle and brought into prohibited areas, thereby preventing safety accidents. In the related art, there is a method of manually inspecting foreign objects under the vehicle, which has low detection efficiency, is prone to missed detections, and has low detection accuracy and precision. Summary of the invention

[0003] The present application aims to solve at least one of the technical problems existing in the related art. To this end, the present application proposes a real-time detection method and device for undercarriage anomalies, which can effectively correct images distorted by other factors such as vehicle speed, and can collect undercarriage images without obvious differences, thereby identifying abnormal conditions existing under the vehicle based on the target undercarriage images, improving detection efficiency, and improving detection precision and accuracy.

[0004] In a first aspect, the present application provides a method for real-time detection of vehicle bottom abnormalities, the method comprising:

[0005] Acquire the actual speed corresponding to the target vehicle at multiple acquisition moments and the initial vehicle bottom images corresponding to the target vehicle at the multiple acquisition moments; the multiple initial vehicle bottom images correspond one to one with the multiple actual speeds;

[0006] Based on each of the actual speeds, an initial vehicle bottom image corresponding to the actual speed is processed to obtain a target vehicle bottom image corresponding to the target vehicle in a target acquisition period; the target acquisition period includes the multiple acquisition moments;

[0007] The target vehicle bottom image is processed to obtain a vehicle bottom detection result corresponding to the target vehicle.

[0008] According to the real-time vehicle underbody abnormality detection method provided in the embodiment of the present application, by acquiring the actual speed of the vehicle and the initial vehicle underbody image at multiple acquisition times, the initial vehicle underbody image is corrected based on the actual speed to obtain a target vehicle underbody image. The image distorted by other factors such as vehicle speed can be effectively corrected, and a vehicle underbody image with no obvious difference can be acquired, so that abnormal conditions existing in the vehicle underbody can be identified based on the target vehicle underbody image, thereby improving the detection efficiency and improving the precision and accuracy of the detection.

[0009] A real-time vehicle bottom abnormality detection method according to an embodiment of the present application, wherein the initial vehicle bottom image corresponding to the actual speed is processed based on each actual speed to obtain a target vehicle bottom image corresponding to the target vehicle in the target acquisition period, comprises:

[0010] Based on the target speed and each of the actual speeds, an initial vehicle bottom image corresponding to the actual speed is deformed to obtain a plurality of first vehicle bottom images; the plurality of first vehicle bottom images correspond one to one to the plurality of the actual speeds;

[0011] Based on the target speed and each of the actual speeds, a first vehicle bottom image corresponding to the actual speed is processed to acquire the target vehicle bottom image.

[0012] In a method for real-time detection of vehicle bottom abnormalities according to an embodiment of the present application, the initial vehicle bottom image corresponding to the actual speed is processed based on the target speed and each actual speed to obtain multiple first vehicle bottom images, including:

[0013] Acquire the length information corresponding to each of the initial vehicle bottom images based on the difference between each of the actual speeds and the target speed;

[0014] Based on each of the length information, each of the initial vehicle bottom images is compressed or stretched to obtain the multiple first vehicle bottom images.

[0015] In a real-time vehicle bottom abnormality detection method according to an embodiment of the present application, the first vehicle bottom image corresponding to the actual speed is processed based on the target speed and each actual speed to obtain the target vehicle bottom image, including:

[0016] Based on the difference between each of the actual speeds and the target speed, a first vehicle bottom image corresponding to the actual speed is cropped to obtain a plurality of second vehicle bottom images; there is no overlapping area between the second vehicle bottom images corresponding to adjacent acquisition moments;

[0017] Based on the acquisition time corresponding to each of the second vehicle bottom images, the plurality of second vehicle bottom images are spliced ​​to obtain the target vehicle bottom image.

[0018] In a real-time vehicle bottom abnormality detection method according to an embodiment of the present application, the processing of the target vehicle bottom image to obtain the vehicle bottom detection result corresponding to the target vehicle includes:

[0019] The target underbody image is input into an underbody anomaly detection model to obtain an underbody detection result corresponding to the target vehicle output by the underbody anomaly detection model; the underbody anomaly detection model is trained using sample underbody images as samples and sample detection results corresponding to the sample underbody images as sample labels.

[0020] A real-time vehicle bottom anomaly detection method according to an embodiment of the present application, wherein the target vehicle bottom image is input into a vehicle bottom anomaly detection model, and a vehicle bottom detection result corresponding to the target vehicle output by the vehicle bottom anomaly detection model is obtained, comprising:

[0021] The target vehicle bottom image is segmented to obtain a plurality of third vehicle bottom images corresponding to the target vehicle; there are overlapping areas between adjacent third vehicle bottom images;

[0022] The plurality of third vehicle underbody images are processed to obtain the vehicle underbody detection result.

[0023] In a method for real-time detection of vehicle bottom abnormalities according to an embodiment of the present application, the processing of the plurality of third vehicle bottom images to obtain the vehicle bottom detection result includes:

[0024] Processing each of the third vehicle bottom images to obtain an initial detection result corresponding to each of the third vehicle bottom images;

[0025] The initial detection results corresponding to the overlapping area are deduplicated, and the multiple initial detection results after deduplication are merged to obtain the vehicle bottom detection result.

[0026] In one embodiment of the present application, a real-time vehicle bottom anomaly detection method is provided, wherein the vehicle bottom anomaly detection model is trained based on the following steps:

[0027] Build an initial detection model;

[0028] Acquire the sample vehicle bottom image and the sample detection result;

[0029] Inputting the sample vehicle bottom image and the sample detection result into the initial detection model to obtain an initial prediction result output by the initial detection model;

[0030] Based on the difference information between the initial prediction result and the sample detection result, the initial detection model is optimized to obtain the underbody abnormality detection model.

[0031] In a real-time vehicle bottom abnormality detection method according to an embodiment of the present application, the initial detection model is optimized based on the difference information between the initial prediction result and the sample detection result to obtain the vehicle bottom abnormality detection model, including:

[0032] Based on the difference information, a hyperparameter optimization algorithm is used to adjust the parameters of the initial detection model to obtain an optimized initial detection model;

[0033] The optimized initial detection model is subjected to asymmetric linear quantization processing to obtain the vehicle bottom anomaly detection model.

[0034] In a second aspect, the present application provides a real-time detection device for vehicle bottom abnormality, comprising:

[0035] A data acquisition module, used to obtain the actual speed corresponding to the target vehicle at multiple acquisition moments, and the initial vehicle bottom images corresponding to the target vehicle at the multiple acquisition moments; the multiple initial vehicle bottom images correspond one to one with the multiple actual speeds;

[0036] An image correction module, configured to process an initial vehicle bottom image corresponding to the actual speed based on each of the actual speeds, and obtain a target vehicle bottom image corresponding to the target vehicle in a target acquisition period; the target acquisition period includes the multiple acquisition moments;

[0037] The abnormality processing module is used to process the target vehicle bottom image and obtain the vehicle bottom detection result corresponding to the target vehicle.

[0038] According to the real-time detection device for vehicle underbody abnormalities provided in the embodiment of the present application, by acquiring the actual speed of the vehicle and the initial vehicle underbody image at multiple acquisition times, the initial vehicle underbody image is corrected based on the actual speed to obtain a target vehicle underbody image. The image distorted by other factors such as vehicle speed can be effectively corrected, and a vehicle underbody image with no obvious difference can be acquired, so that abnormal conditions existing in the vehicle underbody can be identified based on the target vehicle underbody image, thereby improving the detection efficiency and improving the precision and accuracy of the detection.

[0039] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for real-time detection of underbody abnormalities as described in the first aspect above is implemented.

[0040] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for real-time detection of underbody abnormalities as described in the first aspect above.

[0041] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the real-time detection method for underbody abnormalities as described in the first aspect above.

[0042] The above one or more technical solutions in the embodiments of the present application have at least one of the following technical effects:

[0043] By acquiring the actual speed of the vehicle and the initial underbody image at multiple acquisition moments, the initial underbody image is corrected based on the actual speed to obtain a target underbody image. This method can effectively correct the image distorted by other factors such as vehicle speed, and can acquire an underbody image without obvious differences, thereby identifying abnormal conditions under the vehicle based on the target underbody image, thereby improving detection efficiency and improving detection precision and accuracy.

[0044] Furthermore, by correcting the collected underbody image based on the difference information between the actual speed of the vehicle at the current collection moment and the set target speed, a distortion-free target underbody image can be obtained, which effectively reduces the influence of the vehicle speed on the collected underbody image and can collect more underbody images without obvious differences, thereby improving the precision and accuracy of subsequent underbody anomaly detection.

[0045] Furthermore, by modifying the structure of the detection model, the detection model can process large-resolution underbody images; and by lightweighting the detection model, the model's running speed is improved, and the trained underbody anomaly detection model can detect any type of abnormal situation, broadening the applicable scenarios of the underbody anomaly detection model and improving the precision and accuracy of the model's prediction.

[0046] Furthermore, by segmenting the images input to the underbody anomaly detection model, the underbody anomaly detection model can process high-resolution underbody images, retain more detailed features of the underbody images during the detection process, and improve the precision and accuracy of the detection.

[0047] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0049] Figure 1 This is one of the flow charts of the real-time detection method of vehicle bottom abnormality provided in the embodiment of the present application;

[0050] Figure 2 This is the second flow chart of the real-time detection method of vehicle bottom abnormality provided in the embodiment of the present application;

[0051] Figure 3 Schematic diagram of the principle of the real-time detection method of vehicle bottom abnormality provided by the embodiment of the present application;

[0052] Figure 4This is the third flow chart of the real-time detection method for vehicle bottom abnormality provided in the embodiment of the present application;

[0053] Figure 5 This is the fourth flow chart of the real-time detection method of vehicle bottom abnormality provided in the embodiment of the present application;

[0054] Figure 6 is a structural schematic diagram of a real-time detection device for vehicle bottom abnormality provided in an embodiment of the present application;

[0055] Figure 7 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0057] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0058] In conjunction with the accompanying drawings, the real-time underbody abnormality detection method, real-time underbody abnormality detection device, electronic device and readable storage medium provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0059] Among them, the real-time detection method for vehicle bottom anomalies can be applied to a terminal, and can be specifically executed by hardware or software in the terminal.

[0060] The terminal includes, but is not limited to, a portable communication device such as a mobile phone or tablet computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad). It should also be understood that in some embodiments, the terminal may not be a portable communication device, but a desktop computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad).

[0061] In the following various embodiments, a terminal including a display and a touch-sensitive surface is described. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, a mouse and a joystick.

[0062] The embodiment of the present application provides a method for real-time detection of underbody abnormalities. The execution subject of the method can be an electronic device or a functional module or functional entity in the electronic device that can implement the method for real-time detection of underbody abnormalities. The electronic devices mentioned in the embodiment of the present application include but are not limited to mobile phones, tablet computers, computers, cameras, and wearable devices. The method for real-time detection of underbody abnormalities provided in the embodiment of the present application is described below using an electronic device as an example of the execution subject.

[0063] like Figure 1 As shown, the real-time detection method for vehicle bottom abnormality includes: step 110, step 120 and step 130.

[0064] Step 110, obtaining the actual speed corresponding to the target vehicle at multiple acquisition moments, and the initial vehicle bottom image corresponding to the target vehicle at multiple acquisition moments;

[0065] In this step, the target vehicle may be collected after it enters the detection area, wherein the speed and the bottom image of the target vehicle may be collected.

[0066] The multiple collection moments may be consecutive moments, or may be collection moments set based on fixed intervals.

[0067] For example, a collection frequency may be set, and multiple collection times may be generated based on the collection frequency, and the target vehicle may be collected at each collection time.

[0068] The actual speed of the target vehicle is the driving speed of the target vehicle at the time of collection.

[0069] For example, a speedometer may be set in the collection area. After a vehicle enters the collection area, the speedometer is turned on to measure the actual speed of a certain portion of the collected vehicle bottom image.

[0070] The initial vehicle bottom image is used to characterize the state of the vehicle bottom at the corresponding acquisition time, and the initial vehicle bottom image can be acquired by an image acquisition device.

[0071] The image acquisition device may include a camera.

[0072] The multiple initial vehicle bottom images correspond to the multiple actual speeds one by one, that is, each acquisition moment corresponds to an initial vehicle bottom image and an actual speed.

[0073] During the actual implementation process, the standard vehicle speed can be set to Vm / s, the acquisition frequency can be set to Ffps, and the vehicle length passing through the underbody acquisition system can be set to Sm. For example, the standard vehicle speed can be set to 36km / h, the standard image acquisition frequency can be set to 30fps, and the acquired vehicle length can be set to 10m. The error between the image captured by the test camera and the tester can be used for subsequent synchronization of the actual speed and the initial underbody image, wherein (S / V)*F initial underbody images can be acquired.

[0074] like Figure 2 As shown, when the vehicle enters the image acquisition range with an initial speed of 10m / s and an incremental acceleration of 6.25m / s^2, the photoelectric switch can feed back the switch value to the microcontroller, and the microcontroller outputs a control signal to the speedometer and the camera, and controls the camera and the speedometer to turn on. The speed sensor measures the vehicle speed and the bottom image at the current acquisition time in real time. After the vehicle passes through the image acquisition range, 24 initial bottom images can be acquired.

[0075] Step 120: Processing the initial vehicle bottom image corresponding to the actual speed based on each actual speed to obtain a target vehicle bottom image corresponding to the target vehicle in the target acquisition period;

[0076] In this step, the target vehicle has an actual speed and an initial vehicle bottom image at a collection time. The initial vehicle bottom image at the same collection time can be corrected based on the actual speed to obtain the target vehicle bottom image.

[0077] The target collection period includes multiple collection moments, and the target collection period is the period from when the target vehicle enters the collection area to when the target vehicle leaves the collection area.

[0078] like Figure 3 The target vehicle bottom image includes 1 / n of each initial vehicle bottom image, and the target vehicle bottom image can be used to characterize the complete vehicle bottom condition of the target vehicle. For example, by detecting the target vehicle bottom image, it can be determined whether there are foreign objects under the target vehicle.

[0079] Step 130: Process the target vehicle bottom image to obtain a vehicle bottom detection result corresponding to the target vehicle.

[0080] In this step, the underbody detection result may include whether there is a foreign object under the target vehicle, and if there is a foreign object under the vehicle, the underbody detection result includes information such as the location, size, and type of the foreign object.

[0081] The target vehicle bottom image may be subjected to image detection processing to obtain a vehicle bottom detection result. For example, the target vehicle bottom image may be input into a trained detection model to obtain a vehicle bottom detection result output by the detection model.

[0082] During the research and development process, the inventors discovered that in the related technologies, when a vehicle passes through the collection area at a non-uniform speed or does not pass through the collection area at the set speed, the collected image of the vehicle bottom will be distorted, making it difficult to detect foreign objects through the distorted image of the vehicle bottom, thereby reducing the accuracy and precision of the detection.

[0083] According to the real-time vehicle underbody abnormality detection method provided in the embodiment of the present application, by acquiring the actual speed of the vehicle and the initial vehicle underbody image at multiple acquisition times, the initial vehicle underbody image is corrected based on the actual speed to obtain a target vehicle underbody image. The image distorted by other factors such as vehicle speed can be effectively corrected, and a vehicle underbody image with no obvious difference can be acquired, so that abnormal conditions existing in the vehicle underbody can be identified based on the target vehicle underbody image, thereby improving the detection efficiency and improving the precision and accuracy of the detection.

[0084] In some embodiments, step 120 may include:

[0085] Based on the target speed and each actual speed, an initial vehicle bottom image corresponding to the actual speed is deformed to obtain a plurality of first vehicle bottom images;

[0086] The first vehicle bottom image corresponding to the actual speed is processed based on the target speed and each actual speed to obtain a target vehicle bottom image.

[0087] In this embodiment, the target speed is a standard speed of the vehicle, which can be set based on the length of the vehicle, etc. For example, the target speed can be 35km / h or 36km / h, or it can be other values, which can be customized by the user and is not limited in this application.

[0088] The first vehicle bottom image is an image obtained after deforming the initial vehicle bottom image. Multiple first vehicle bottom images correspond to multiple actual speeds one by one, that is, at the same acquisition time, the target vehicle corresponds to one first vehicle bottom image and one actual speed.

[0089] Then, the first vehicle bottom image may be processed based on the actual speed and the target speed. For example, a plurality of first vehicle bottom images may be spliced ​​to obtain a target vehicle bottom image.

[0090] In some embodiments, processing the initial vehicle bottom image corresponding to the actual speed based on the target speed and each actual speed to obtain multiple first vehicle bottom images may include:

[0091] Acquire the length information corresponding to each initial vehicle bottom image based on the difference between each actual speed and the target speed;

[0092] Based on each length information, each initial vehicle bottom image is compressed or stretched to obtain a plurality of first vehicle bottom images.

[0093] In this embodiment, the difference between the actual speed and the target speed may include a difference between the two, or may also include a ratio between the two, etc.

[0094] The initial vehicle bottom image needs to be deformed to the calculated length to obtain the first vehicle bottom image.

[0095] For example, based on the difference between the actual speed and the target speed, length information may be obtained, and then the initial vehicle bottom image is compressed or stretched to the length information to obtain the first vehicle bottom image.

[0096] In the actual implementation process, the initial vehicle bottom image can be processed based on the formula:

[0097]

[0098] Wherein, L1 is the length of the first vehicle bottom image, L0 is the length of the initial vehicle bottom image, Va is the actual speed of the target vehicle corresponding to the initial vehicle bottom image, and V is the target speed.

[0099] In some embodiments, processing the first vehicle bottom image corresponding to the actual speed based on the target speed and each actual speed to obtain the target vehicle bottom image may include:

[0100] Based on the difference between each actual speed and the target speed, a first vehicle bottom image corresponding to the actual speed is cropped to obtain a plurality of second vehicle bottom images;

[0101] Based on the acquisition time corresponding to each second vehicle bottom image, multiple second vehicle bottom images are stitched together to obtain a target vehicle bottom image.

[0102] In this embodiment, there is an overlapping area between the first vehicle bottom images corresponding to adjacent acquisition moments.

[0103] Based on the difference between the actual speed and the target speed, the first vehicle bottom images may be collected and processed to remove overlapping areas between adjacent first vehicle bottom images.

[0104] There is no overlapping area between the second vehicle bottom images corresponding to adjacent acquisition moments.

[0105] Based on the acquisition time corresponding to each second vehicle bottom image, a plurality of second vehicle bottom images are sequentially stitched together to obtain a target vehicle bottom image including the complete vehicle bottom.

[0106] In the actual implementation process, the first vehicle bottom image can be processed based on the formula:

[0107]

[0108] Wherein, L is the length of the second vehicle bottom image, L1 is the length of the first vehicle bottom image, Va is the actual speed of the target vehicle corresponding to the first vehicle bottom image, and V is the target speed.

[0109] According to the real-time underbody abnormality detection method provided in the embodiment of the present application, the collected underbody image is corrected by using the difference information between the actual speed of the vehicle at the current collection moment and the set target speed, so that a distortion-free target underbody image can be obtained, thereby effectively reducing the influence of the vehicle speed on the collected underbody image, and being able to collect more underbody images without obvious differences, thereby improving the precision and accuracy of subsequent underbody abnormality detection.

[0110] In some embodiments, step 130 may include:

[0111] The target vehicle underbody image is input into the vehicle underbody anomaly detection model to obtain the vehicle underbody detection result corresponding to the target vehicle output by the vehicle underbody anomaly detection model.

[0112] In this embodiment, the vehicle bottom anomaly detection model can be trained using sample vehicle bottom images as samples and sample detection results corresponding to the sample vehicle bottom images as sample labels.

[0113] The sample vehicle bottom images include normal vehicle bottom images and abnormal vehicle bottom images.

[0114] The sample vehicle bottom image can be acquired based on step 110 and step 120 . The sample vehicle bottom image includes complete vehicle bottom features, and the sample vehicle bottom image is an image without distortion.

[0115] The sample detection results may be obtained based on manual annotation on the sample vehicle bottom image, or may be obtained based on other methods, which is not limited in this application.

[0116] like Figure 4 As shown, in some embodiments, the underbody anomaly detection model can be trained based on the following steps:

[0117] Build an initial detection model;

[0118] Obtain sample vehicle bottom images and sample detection results;

[0119] Input the sample vehicle bottom image and the sample detection result into the initial detection model to obtain the initial prediction result output by the initial detection model;

[0120] Based on the difference information between the initial prediction results and the sample detection results, the initial detection model is optimized to obtain the underbody anomaly detection model.

[0121] In this embodiment, the sample vehicle bottom image may be acquired based on step 110 and step 120 .

[0122] In the actual implementation process, a large number of vehicle bottom images can be acquired based on step 110 and step 120 to construct a normal image data set, and then foreign objects are added under the vehicle to collect abnormal vehicle bottom images.

[0123] Then the normal images are divided into training set and test set in the ratio of 8:2, and then the abnormal underbody images are added to the test set to test whether the anomaly can be identified.

[0124] The initial detection model may include a feature extraction module, a feature library, and a detection algorithm.

[0125] The feature extraction module may include a plurality of sequentially connected learnable residual blocks (ReZero Block).

[0126] Among them, ReZero Block can include convolutional kernel (Convolutional Layer), batch normalization layer (Batch Normalization, BN), non-linear activation function (Gaussian Error Linear Unit, GELU) and learnable residual connection (ReZero).

[0127] The size of the convolution kernel can be 7x7, or other sizes, which is not limited in this application.

[0128] Setting 7x7 Conv can provide a larger receptive field for capturing global features or larger structural information in the image, and can produce fewer features for larger input images.

[0129] The ReZero connection is a learnable parameter α, which is multiplied by the residual (i.e., the difference between the input and the output after convolution+BN+GELU), and then the result is added to the main path, so that the initial detection model can dynamically adjust the proportion of the residual during training.

[0130] The feature library is used to store the features of the normal vehicle bottom image. The vector features of the normal vehicle bottom image are extracted by the feature extraction module. To reduce the size of the feature library, the vector features in the feature library can be sampled and approximated.

[0131] The detection algorithm can be based on the above-mentioned feature library and detect anomalies through the nearest neighbor method. First, the features of the model input image can be obtained through the feature extraction module, and then the distance between the input image features and the anomalies in the feature library can be calculated by the nearest neighbor method. When the distance exceeds a certain threshold, it can be considered that there is an anomaly in the input image.

[0132] The input size of the initial detection model can be 1000x1000.

[0133] Based on the difference information between the initial prediction results and the sample detection results, the parameters of the initial detection model can be adjusted to optimize the initial detection model.

[0134] In some embodiments, based on the difference information between the initial prediction result and the sample detection result, optimizing the initial detection model to obtain the underbody abnormality detection model may include:

[0135] Based on the difference information, the hyperparameter optimization algorithm is used to adjust the parameters of the initial detection model to obtain the optimized initial detection model;

[0136] The optimized initial detection model is subjected to asymmetric linear quantization processing to obtain the underbody anomaly detection model.

[0137] In this embodiment, the hyperparameters of the initial detection model and the structural parameters of the neural network (such as the depth of the neural network, the learning rate, etc.) can be searched through a hyperparameter optimization algorithm (Tree-structured Parzen Estimator, TPE).

[0138] The hyperparameter optimization algorithm is a variant of the Bayesian optimization algorithm.

[0139] The hyperparameter optimization algorithm can divide the hyperparameter space into two parts: suggestions and observations. The suggestion part is a probability distribution used to generate new hyperparameter values; the observation part is the hyperparameter value that has been evaluated. The hyperparameter optimization algorithm can select the next hyperparameter value by comparing the probability density functions of the suggestion part and the observation part.

[0140] The optimized initial detection model can be subjected to asymmetric linear quantization processing, and the parameters of the initial detection model can be quantized from fp32 to int8 to improve the model operation efficiency and realize real-time detection.

[0141] In the actual execution process, a calibration data set can be obtained through steps 110 and 120; then the model is run using the calibration data set to obtain statistical information (such as the distribution of activation values) for determining the best quantization parameters; then the quantization parameters can be applied to convert the FP32 parameters of the model into INT8 parameters.

[0142] According to the real-time underbody anomaly detection method provided in the embodiment of the present application, by modifying the structure of the detection model, the detection model can process underbody images with large resolution; and by lightweight processing of the detection model, the running speed of the model is improved, and the trained underbody anomaly detection model can detect any type of abnormal situation, thereby broadening the applicable scenarios of the underbody anomaly detection model and improving the precision and accuracy of the model prediction.

[0143] like Figure 5 As shown, in some embodiments, inputting the target vehicle bottom image into the vehicle bottom anomaly detection model, and obtaining the vehicle bottom detection result corresponding to the target vehicle output by the vehicle bottom anomaly detection model may include:

[0144] Segmenting the target vehicle bottom image to obtain a plurality of third vehicle bottom images corresponding to the target vehicle;

[0145] The plurality of third vehicle bottom images are processed to obtain a vehicle bottom detection result.

[0146] In this embodiment, the target vehicle bottom image may be divided into a plurality of overlapping images, namely, the third vehicle bottom image, by means of a sliding window.

[0147] There is an overlapping area between adjacent third vehicle bottom images.

[0148] A plurality of third underbody images may be input into the underbody anomaly detection model at one time, and processing results corresponding to each third underbody image output by the underbody anomaly detection model may be obtained respectively, thereby obtaining an underbody detection result.

[0149] In some embodiments, processing a plurality of third vehicle bottom images to obtain a vehicle bottom detection result may include:

[0150] Processing each third vehicle bottom image to obtain an initial detection result corresponding to each third vehicle bottom image;

[0151] The initial detection results corresponding to the overlapping areas are deduplicated, and multiple initial detection results after deduplication are merged to obtain the vehicle bottom detection result.

[0152] In this embodiment, the initial detection result is the detection result corresponding to the third vehicle underbody image. There may be overlapping areas between two adjacent third vehicle underbody images, and there may also be repeated detection results in the initial detection results corresponding to two adjacent third vehicle underbody images.

[0153] The detection results corresponding to the overlapping areas may be deduplicated, and then the multiple initial detection results after deduplication may be merged.

[0154] like Figure 5As shown, in the actual execution process, for example, when the size of the target vehicle bottom image is 12000x2000, the target vehicle bottom image can be divided into multiple third vehicle bottom images with overlapping areas by means of a sliding window. When the sliding step size is 500, 33 third vehicle bottom images can be obtained.

[0155] Then, the multiple third underbody images can be pre-processed in batches, including normalization and other operations, and then the multiple third underbody images can be input into the underbody anomaly model at one time to obtain multiple initial detection results output by the underbody anomaly detection model.

[0156] After the vehicle bottom anomaly detection model inference is completed, the detection results of multiple third vehicle bottom images need to be merged.

[0157] When the 33 images are input into the underbody anomaly detection model, the processing results of the 33 images need to be merged. For the detection results of the overlapping areas, the maximum area method can be used to remove duplicates, and then the final results can be output.

[0158] According to the real-time underbody anomaly detection method provided in the embodiment of the present application, by segmenting the image input to the underbody anomaly detection model, the underbody anomaly detection model can process high-resolution underbody images, and more detailed features of the underbody images can be retained during the detection process, thereby improving the precision and accuracy of the detection.

[0159] In some embodiments, after step 130, the method for real-time detection of vehicle bottom abnormality may further include:

[0160] When it is determined based on the under-vehicle detection result that there is an abnormality under the vehicle corresponding to the target vehicle, an alarm message is output.

[0161] In this embodiment, the alarm information may be in the form of voice or flashing lights, or may be in other forms, which are not limited in this application.

[0162] After the user receives the alarm information, he can check the target vehicle and perform other operations.

[0163] The real-time detection device for vehicle underbody abnormality provided in the present application is described below. The real-time detection device for vehicle underbody abnormality described below and the real-time detection method for vehicle underbody abnormality described above can be referenced to each other.

[0164] The real-time detection method for underbody abnormality provided in the embodiment of the present application can be executed by a real-time detection device for underbody abnormality. In the embodiment of the present application, the real-time detection method for underbody abnormality performed by the real-time detection device for underbody abnormality is taken as an example to illustrate the real-time detection device for underbody abnormality provided in the embodiment of the present application.

[0165] An embodiment of the present application also provides a real-time detection device for vehicle bottom abnormalities.

[0166] like Figure 6 As shown, the real-time detection device for vehicle bottom abnormality includes: a data acquisition module 610, an image correction module 620 and an abnormality processing module 630.

[0167] The data acquisition module 610 is used to obtain the actual speed corresponding to the target vehicle at multiple acquisition times and the initial vehicle bottom images corresponding to the target vehicle at multiple acquisition times; the multiple initial vehicle bottom images correspond to the multiple actual speeds one by one;

[0168] The image correction module 620 is used to process the initial vehicle bottom image corresponding to the actual speed based on each actual speed, and obtain the target vehicle bottom image corresponding to the target vehicle in the target acquisition period; the target acquisition period includes multiple acquisition moments;

[0169] The abnormality processing module 630 is used to process the target vehicle bottom image and obtain the vehicle bottom detection result corresponding to the target vehicle.

[0170] According to the real-time detection device for vehicle underbody abnormalities provided in the embodiment of the present application, by acquiring the actual speed of the vehicle and the initial vehicle underbody image at multiple acquisition times, the initial vehicle underbody image is corrected based on the actual speed to obtain a target vehicle underbody image. The image distorted by other factors such as vehicle speed can be effectively corrected, and a vehicle underbody image with no obvious difference can be acquired, so that abnormal conditions existing in the vehicle underbody can be identified based on the target vehicle underbody image, thereby improving the detection efficiency and improving the precision and accuracy of the detection.

[0171] In some embodiments, the image correction module 620 may also be used to:

[0172] Based on the target speed and each of the actual speeds, an initial vehicle bottom image corresponding to the actual speed is deformed to obtain a plurality of first vehicle bottom images; the plurality of first vehicle bottom images correspond one to one to the plurality of the actual speeds;

[0173] Based on the target speed and each of the actual speeds, a first vehicle bottom image corresponding to the actual speed is processed to acquire the target vehicle bottom image.

[0174] In some embodiments, the image correction module 620 may also be used to:

[0175] Acquire the length information corresponding to each of the initial vehicle bottom images based on the difference between each of the actual speeds and the target speed;

[0176] Based on each of the length information, each of the initial vehicle bottom images is compressed or stretched to obtain the multiple first vehicle bottom images.

[0177] In some embodiments, the image correction module 620 may also be used to:

[0178] Based on the difference between each of the actual speeds and the target speed, a first vehicle bottom image corresponding to the actual speed is cropped to obtain a plurality of second vehicle bottom images; there is no overlapping area between the second vehicle bottom images corresponding to adjacent acquisition moments;

[0179] Based on the acquisition time corresponding to each of the second vehicle bottom images, the plurality of second vehicle bottom images are spliced ​​to obtain the target vehicle bottom image.

[0180] In some embodiments, the exception handling module 630 may also be used to:

[0181] The target underbody image is input into an underbody anomaly detection model to obtain an underbody detection result corresponding to the target vehicle output by the underbody anomaly detection model; the underbody anomaly detection model is trained using sample underbody images as samples and sample detection results corresponding to the sample underbody images as sample labels.

[0182] In some embodiments, the exception handling module 630 may also be used to:

[0183] The target vehicle bottom image is segmented to obtain a plurality of third vehicle bottom images corresponding to the target vehicle; there are overlapping areas between adjacent third vehicle bottom images;

[0184] The plurality of third vehicle underbody images are processed to obtain the vehicle underbody detection result.

[0185] In some embodiments, the exception handling module 630 may also be used to:

[0186] Processing each of the third vehicle bottom images to obtain an initial detection result corresponding to each of the third vehicle bottom images;

[0187] The initial detection results corresponding to the overlapping area are deduplicated, and the multiple initial detection results after deduplication are merged to obtain the vehicle bottom detection result.

[0188] In some embodiments, the anomaly processing module 630 may also be used to train the vehicle bottom anomaly detection model based on the following steps:

[0189] Build an initial detection model;

[0190] Acquire the sample vehicle bottom image and the sample detection result;

[0191] Inputting the sample vehicle bottom image and the sample detection result into the initial detection model to obtain an initial prediction result output by the initial detection model;

[0192] Based on the difference information between the initial prediction result and the sample detection result, the initial detection model is optimized to obtain the underbody abnormality detection model.

[0193] In some embodiments, the exception handling module 630 may also be used to:

[0194] Based on the difference information, a hyperparameter optimization algorithm is used to adjust the parameters of the initial detection model to obtain an optimized initial detection model;

[0195] The optimized initial detection model is subjected to asymmetric linear quantization processing to obtain the vehicle bottom anomaly detection model.

[0196] The real-time detection device for abnormal undercarriage in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices other than the terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a palmtop computer, a vehicle-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., and the embodiment of the present application is not specifically limited.

[0197] The real-time detection device for vehicle bottom abnormality in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an IOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0198] The real-time detection device for vehicle bottom abnormality provided in the embodiment of the present application can achieve Figures 1 to 5 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0199] In some embodiments, Figure 7As shown, an embodiment of the present application also provides an electronic device 700, including a processor 701, a memory 702, and a computer program stored in the memory 702 and executable on the processor 701. When the program is executed by the processor 701, each process of the above-mentioned embodiment of the real-time detection method for underbody abnormalities is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0200] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0201] On the other hand, the present application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the various processes of the above-mentioned embodiment of the real-time detection method for underbody abnormalities and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0202] On the other hand, the present application also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the various processes of the above-mentioned real-time detection method for underbody abnormalities and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0203] On the other hand, an embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned embodiment of the real-time detection method for underbody abnormalities, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0204] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0205] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0206] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution can be essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time detection method for vehicle bottom abnormality, characterized in that: include: Acquire the actual speed of the target vehicle at multiple acquisition moments and the initial vehicle bottom image of the target vehicle at the multiple acquisition moments; The multiple initial vehicle bottom images correspond one to one with the multiple actual speeds; Based on each of the actual speeds, an initial vehicle bottom image corresponding to the actual speed is processed to obtain a target vehicle bottom image corresponding to the target vehicle in a target acquisition period; The target collection period includes the multiple collection moments; The target vehicle bottom image is processed to obtain a vehicle bottom detection result corresponding to the target vehicle.

2. The method for real-time detection of vehicle bottom abnormality according to claim 1, characterized in that: The processing of the initial vehicle bottom image corresponding to the actual speed based on each actual speed to obtain the target vehicle bottom image corresponding to the target vehicle in the target acquisition period includes: Based on the target speed and each of the actual speeds, an initial vehicle bottom image corresponding to the actual speed is deformed to obtain a plurality of first vehicle bottom images; the plurality of first vehicle bottom images correspond one to one to the plurality of the actual speeds; Based on the target speed and each of the actual speeds, a first vehicle bottom image corresponding to the actual speed is processed to acquire the target vehicle bottom image.

3. The method for real-time detection of vehicle bottom abnormality according to claim 2, characterized in that: The initial vehicle bottom image corresponding to the actual speed is processed based on the target speed and each actual speed to obtain a plurality of first vehicle bottom images, including: Acquire the length information corresponding to each of the initial vehicle bottom images based on the difference between each of the actual speeds and the target speed; Based on each of the length information, each of the initial vehicle bottom images is compressed or stretched to obtain the multiple first vehicle bottom images.

4. The method for real-time detection of vehicle bottom abnormality according to claim 2, characterized in that: The step of processing the first vehicle bottom image corresponding to the actual speed based on the target speed and each actual speed to obtain the target vehicle bottom image includes: Based on the difference between each of the actual speeds and the target speed, a first vehicle bottom image corresponding to the actual speed is cropped to obtain a plurality of second vehicle bottom images; there is no overlapping area between the second vehicle bottom images corresponding to adjacent acquisition moments; Based on the acquisition time corresponding to each of the second vehicle bottom images, the plurality of second vehicle bottom images are spliced ​​to obtain the target vehicle bottom image.

5. The method for real-time detection of vehicle bottom abnormality according to any one of claims 1 to 4, characterized in that: The processing of the target vehicle bottom image to obtain a vehicle bottom detection result corresponding to the target vehicle includes: The target underbody image is input into an underbody anomaly detection model to obtain an underbody detection result corresponding to the target vehicle output by the underbody anomaly detection model; the underbody anomaly detection model is trained using sample underbody images as samples and sample detection results corresponding to the sample underbody images as sample labels.

6. The method for real-time detection of vehicle bottom abnormality according to claim 5, characterized in that: The step of inputting the target vehicle bottom image into a vehicle bottom anomaly detection model and obtaining a vehicle bottom detection result corresponding to the target vehicle output by the vehicle bottom anomaly detection model includes: The target vehicle bottom image is segmented to obtain a plurality of third vehicle bottom images corresponding to the target vehicle; there are overlapping areas between adjacent third vehicle bottom images; The plurality of third vehicle underbody images are processed to obtain the vehicle underbody detection result.

7. The method for real-time detection of vehicle bottom abnormality according to claim 6, characterized in that: The processing of the plurality of third vehicle bottom images to obtain the vehicle bottom detection result includes: Processing each of the third vehicle bottom images to obtain an initial detection result corresponding to each of the third vehicle bottom images; The initial detection results corresponding to the overlapping area are deduplicated, and the multiple initial detection results after deduplication are merged to obtain the vehicle bottom detection result.

8. The method for real-time detection of vehicle bottom abnormality according to claim 5, characterized in that: The vehicle bottom anomaly detection model is trained based on the following steps: Build an initial detection model; Acquire the sample vehicle bottom image and the sample detection result; Inputting the sample vehicle bottom image and the sample detection result into the initial detection model to obtain an initial prediction result output by the initial detection model; Based on the difference information between the initial prediction result and the sample detection result, the initial detection model is optimized to obtain the underbody abnormality detection model.

9. The method for real-time detection of vehicle bottom abnormality according to claim 8, characterized in that: The optimizing the initial detection model based on the difference information between the initial prediction result and the sample detection result to obtain the vehicle bottom abnormality detection model includes: Based on the difference information, a hyperparameter optimization algorithm is used to adjust the parameters of the initial detection model to obtain an optimized initial detection model; The optimized initial detection model is subjected to asymmetric linear quantization processing to obtain the vehicle bottom abnormality detection model.

10. A real-time detection device for vehicle bottom abnormality, characterized in that: include: A data acquisition module, used to acquire the actual speed of the target vehicle at multiple acquisition moments and the initial vehicle bottom image corresponding to the target vehicle at the multiple acquisition moments; The multiple initial vehicle bottom images correspond one to one with the multiple actual speeds; An image correction module is used to process the initial vehicle bottom image corresponding to the actual speed based on each actual speed, and obtain a target vehicle bottom image corresponding to the target vehicle in the target acquisition period; The target collection period includes the multiple collection moments; The abnormality processing module is used to process the target vehicle bottom image and obtain the vehicle bottom detection result corresponding to the target vehicle.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the real-time detection method for vehicle bottom abnormality as described in any one of claims 1 to 9 is implemented.

12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the real-time detection method for vehicle bottom abnormality as described in any one of claims 1 to 9 is implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the real-time detection method for vehicle bottom abnormality as described in any one of claims 1 to 9 is implemented.

Images