Vehicle hazard light recognition model construction, recognition method, device and medium

Abstract

This invention relates to the field of autonomous driving, specifically to a method, apparatus, and medium for constructing and recognizing a vehicle hazard light recognition model. The method includes: acquiring continuous frame image data of the vehicle's hazard lights being on or off to construct a sample set; training a temporal network model using the sample set to obtain a vehicle hazard light recognition model. By implementing this invention, continuous frame image data is used as samples, and these samples are used to train a temporal network model containing convolutional neural networks, self-attention mechanisms, and recurrent neural networks to obtain the vehicle hazard light recognition model. This method obtains a vehicle hazard light recognition model by training a temporal network model. When performing temporal analysis on the data, the temporal network can effectively resist changes and interference from environmental noise factors such as light and distance, exhibiting good robustness. Therefore, it solves the problem that existing hazard light recognition methods are sensitive to external environments, affecting recognition accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic driving, in particular to a vehicle double flash signal lamp recognition model construction method, device and medium. BACKGROUND

[0002] The double flash signal lamp is a common warning lamp on a car. For an automatic driving or intelligent vehicle, the double flash signal lamp of other vehicles is recognized to determine the driving intention of the front or rear vehicle, so as to adjust the driving state of the vehicle itself, thereby improving the driving safety and intelligent level of the automatic driving vehicle.

[0003] At present, the double flash signal lamp recognition firstly determines the taillight region through HSV color space threshold filtering, Gaussian transformation and normalization operation; then the maximum inter-class variance method and morphological transformation method are used to obtain more accurate taillight region and its contour; finally, the opening and extinguishing time of the vehicle taillight is tracked and recorded in real time, and whether the double flash signal lamp is opened is determined according to the opening rule of the taillight. The method uses traditional image processing method, which is sensitive to light, deformation scene and real-time performance. SUMMARY

[0004] Therefore, the present application provides a vehicle double flash signal lamp recognition model construction method, device and medium to solve the problem that the existing double flash signal lamp recognition is sensitive to the external environment and affects the recognition accuracy.

[0005] In a first aspect, the present application provides a vehicle double flash signal lamp recognition model construction method, which comprises: acquiring a continuous frame image data set of a vehicle tail double flash signal lamp opening or closing, and setting a label of the double flash signal lamp opening or closing for the image data; training a time sequence network model by using the sample set to obtain a vehicle double flash signal lamp recognition model, wherein the time sequence network model comprises a convolutional neural network, a self-attention mechanism network and a recurrent neural network.

[0006] The vehicle double flash signal lamp recognition model construction method provided by the present application uses continuous frame image data as a sample, trains a time sequence network model comprising a convolutional neural network, a self-attention mechanism network and a recurrent neural network by using the sample, and obtains a vehicle double flash signal lamp recognition model. The method trains a vehicle double flash signal lamp recognition model by training a time sequence network model, and the time sequence network can effectively resist the changes and interference of environmental noise factors such as light and distance when analyzing the time sequence of data, and has good robustness. Thus, the problem that the existing double flash signal lamp recognition is sensitive to the external environment and affects the recognition accuracy is solved.

[0007] In an optional implementation, the method for constructing a sample set by acquiring continuous frame image data of a vehicle tail double flash signal lamp being turned on or off, and setting a label of the double flash signal lamp being turned on or off for the image data, comprises: acquiring video data of the vehicle tail double flash signal lamp being turned on and off; dividing the video data according to a preset time interval; obtaining a plurality of continuous frame image data; and setting a label of the double flash signal lamp being turned on or off for each continuous frame image data to form the sample set.

[0008] The method for constructing a vehicle double flash signal lamp recognition model provided by the embodiment of the application realizes construction of a sample set by dividing video data to obtain continuous frame image data.

[0009] In an optional implementation, the recurrent neural network is a long short-term memory network, the sample set is used to train a time sequence network model to obtain a vehicle double flash signal lamp recognition model, which comprises: inputting preset frame image data in the sample set into the time sequence network model, using a convolutional neural network to extract features of each frame image, inputting the extracted features into a self-attention mechanism network to calculate similarities between the features, inputting features corresponding to a plurality of frames and having a similarity greater than a threshold into a long short-term memory network, and outputting a detection result of whether the double flash signal lamp is turned on; updating parameters of the time sequence network model based on a difference between the detection result and the label; and repeating the steps of detection and parameter updating by using the time sequence network model until a preset training number of rounds is reached.

[0010] The method for constructing a vehicle double flash signal lamp recognition model provided by the embodiment of the application uses a convolutional neural network to extract features of each frame image, sets a self-attention mechanism network, so that the time sequence network model pays more attention to features that are more conducive to double flash signal lamp recognition and ignores features that are less important. Then, a long short-term memory network is used to process a feature sequence composed of the features to realize detection of whether continuous frame image data contains a turned-on double flash signal lamp.

[0011] In an optional implementation, the parameters of the time sequence network model are updated based on a difference between the detection result and the label, which comprises: determining a loss function based on the difference between the detection result and the label; and updating the parameters of the time sequence network model by using a gradient descent method based on the loss function.

[0012] In a second aspect, the present application provides a vehicle double flash signal lamp recognition model construction device, the device comprising: a sample set construction module configured to obtain a sample set of continuous frame image data of a vehicle tail double flash signal lamp being turned on or off, the image data being provided with a label of the double flash signal lamp being turned on or off; and a model construction module configured to train a time sequence network model using the sample set to obtain a vehicle double flash signal lamp recognition model, the time sequence network model comprising a convolutional neural network, a self-attention mechanism network, and a recurrent neural network.

[0013] In a third aspect, the present application provides a vehicle double flash signal lamp recognition method, the method comprising: obtaining continuous multiple frames of vehicle tail image data; inputting the multiple frames of vehicle tail image data into a vehicle double flash signal lamp recognition model constructed by the vehicle double flash signal lamp recognition model construction method of the first aspect or any of the corresponding embodiments thereof to obtain a recognition result of whether the double flash signal lamp is turned on.

[0014] The vehicle double flash signal lamp recognition method provided by the embodiments of the present application can obtain the recognition result of whether the double flash signal lamp is turned on by inputting the continuous multiple frames of vehicle tail image data into the vehicle double flash signal lamp recognition model. Since the vehicle double flash signal lamp recognition model is trained using a time sequence network model, and the state of the double flash signal lamp is flickering, the time sequence network can well capture the time sequence, and thus the recognition method can accurately recognize whether the double flash signal lamp is turned on.

[0015] In a fourth aspect, the present application provides a vehicle double flash signal lamp recognition device, the device comprising: a data acquisition module configured to obtain continuous multiple frames of vehicle tail image data; and a recognition module configured to input the vehicle tail image data into a vehicle double flash signal lamp recognition model constructed by the vehicle double flash signal lamp recognition model construction method of the first aspect or any of the corresponding embodiments thereof to obtain a recognition result of whether the double flash signal lamp is turned on.

[0016] In a fifth aspect, the present application provides a computer device, comprising: a memory and a processor, which are communicatively connected to each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the vehicle double flash signal lamp recognition model construction method of the first aspect or any of the corresponding embodiments thereof and the vehicle double flash signal lamp recognition method of the third aspect.

[0017] In a sixth aspect, the present application provides a computer readable storage medium, which stores computer instructions for causing a computer to execute the vehicle double flash signal lamp recognition model construction method of the first aspect or any of the corresponding embodiments thereof and the vehicle double flash signal lamp recognition method of the third aspect. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the specific embodiments of the present application or the prior art, the drawings required to be used in the description of the specific embodiments or the prior art will be briefly introduced. Obviously, the drawings described below are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0019] Figure 1 is a flowchart of a vehicle double flash signal lamp recognition model construction method according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of a vehicle double flash signal lamp recognition model construction principle when there is an occlusion according to an embodiment of the present application;

[0021] Figure 3 is a flowchart of a vehicle double flash signal lamp recognition method according to an embodiment of the present application;

[0022] Figure 4 is a structural block diagram of a vehicle double flash signal lamp recognition model construction device according to an embodiment of the present application;

[0023] Figure 5 is a structural block diagram of a vehicle double flash signal lamp recognition device according to an embodiment of the present application;

[0024] Figure 6 is a hardware structure schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0026] According to an embodiment of the present application, a vehicle double flash signal lamp recognition model construction method embodiment is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] In the present embodiment, a vehicle double flash signal lamp recognition model construction method is provided, which can be used in electronic devices such as computers or mobile terminals, etc. Figure 1is a flowchart of a vehicle double flash signal lamp recognition model construction method according to an embodiment of the present application, as shown in the figure, the flow includes the following steps: Figure 1

[0028] In step S101, a sample set is constructed by acquiring continuous frame image data of the opening or closing of the vehicle tail double flash signal lamp, and the image data is provided with a label of the opening or closing of the double flash signal lamp. Specifically, when the vehicle opens the double flash signal lamp, the double flash signal lamp will alternately emit light and extinguish within a certain time interval. If there is only a single frame of image, there is only the state of the double flash signal lamp being on or off, and it is impossible to accurately reflect whether the double flash signal lamp is on. Therefore, continuous frame image data is used as each sample in the sample set. When the sample is collected when the double flash signal lamp is on, the state of the double flash signal lamp in the continuous frame image is alternately emitting light and extinguishing, and the sample is assigned a label of the double flash signal lamp being on; when the sample is collected when the double flash signal lamp is off, the state of the double flash signal lamp in the continuous frame image is always extinguished, and the sample is assigned a label of the double flash signal lamp being off.

[0029] For example, when the continuous frame image data includes T frame image data, each sample in the sample set can be represented as x, the size of which is represented as (T, W, H, C), and the corresponding label is represented as c (in lowercase). T represents continuous T frame images, the width and height of each image frame are W and H respectively, and C (in uppercase) is the number of channels of the image.

[0030] In step S102, the sample set is used to train a time sequence network model to obtain a vehicle double flash signal lamp recognition model, and the time sequence network model includes a convolutional neural network, a self-attention mechanism network and a recurrent neural network. In the specific training, the sample set can be divided into a training set and a test set, the training set is used to train the time sequence network model, the training set can be input into the time sequence network model, the parameters of the time sequence network model are adjusted through the relationship between the output and the corresponding label, and then the training of the time sequence network model is completed through an iterative process. The convolutional neural network and the self-attention mechanism network can only process single frame images, so the recurrent neural network (RNN) is added to process the image feature sequence (i.e. the sequence formed by the features extracted from the continuous frame images). In addition, after the training is completed, the vehicle double flash signal lamp recognition model obtained by the training can also be tested by using the test set to see whether it meets the preset requirements. When the preset requirements are not met, the training can be performed again until the vehicle double flash signal lamp recognition model obtained by the training meets the preset requirements.

[0031] ​The vehicle double flash signal lamp recognition model construction method provided by the embodiment of the present application adopts continuous frame image data as samples, trains a time sequence network model containing a convolutional neural network, a self-attention mechanism network and a recurrent neural network using the samples, and obtains a vehicle double flash signal lamp recognition model. The method trains a vehicle double flash signal lamp recognition model through the time sequence network model, and the time sequence network can effectively resist changes and interference of environmental noise factors such as light and distance when performing time sequence analysis on data, and has good robustness. Thus, the problem that the existing double flash signal lamp recognition is sensitive to external environment and affects recognition accuracy is solved.

[0032] A vehicle double flash signal lamp recognition model construction method is provided in the embodiment, and the method comprises the following steps:

[0033] In step S201, continuous frame image data of a vehicle tail double flash signal lamp being turned on or off is obtained to construct a sample set, and the image data is provided with a label of the double flash signal lamp being turned on or off.

[0034] Specifically, step S201 comprises the following steps:

[0035] In step S2011, video data of the vehicle tail double flash signal lamp being turned on and off is obtained. When the video data is obtained, the video data of the vehicle tail when the double flash signal lamp of different vehicle models is turned on and off can be recorded. Alternatively, the video data can be obtained through a video data set of the vehicle tail double flash signal lamp being turned on and off disclosed in related technologies, and the source of the video data is not limited in the embodiment of the present application.

[0036] In step S2012, the video data is divided according to a preset time interval, and a plurality of continuous frame image data is obtained. When the video data is divided, the video data is divided according to a preset time interval, for example, a time interval T, and a plurality of continuous frame image data is obtained, and each continuous frame image data comprises T continuous frame image data.

[0037] In step S2013, a label of the double flash signal lamp being turned on or off is set for each continuous frame image data to form a sample set. Specifically, each continuous frame image data is obtained by dividing the video data, and the video data is the video data when the double flash signal lamp is turned on or off. Therefore, a label of the double flash signal lamp being turned on or off can be set for each continuous frame image according to the state of the double flash signal lamp being turned on or off in the video data before division.

[0038] In step S202, the sample set is used to train a time sequence network model, and a vehicle double flash signal lamp recognition model is obtained, wherein the time sequence network model comprises a convolutional neural network, a self-attention mechanism network and a recurrent neural network.

[0039] Specifically, the recurrent neural network is a long short-term memory network, and the step S202 includes:

[0040] In step S2021, the preset frame image data in the sample set is input into a time sequence network model, a convolutional neural network is used to extract features of each frame of image, the extracted features are input into a self-attention mechanism network to calculate the similarity between the features, features with a similarity greater than a threshold value corresponding to multiple frames are input into a long short-term memory network, and a detection result of whether the double flash signal lamp is turned on is output. The convolutional neural network can adopt a network structure such as VGG or ResNet, and the convolutional neural network can realize extraction of features such as edges, colors and textures of objects in each frame of image. The features extracted by the convolutional neural network can be input into the self-attention mechanism network, the self-attention mechanism network calculates the similarity or correlation between the input features, and the self-attention mechanism network can make the time sequence network model pay more attention to features more conducive to double flash signal lamp recognition and ignore unimportant features. Then the features processed by the self-attention mechanism network are input into the long short-term memory network (LSTM, Long Short-Term Memory), and at each time step, the LSTM network calculates the input data and the memory state of the previous time step, and then updates the internal state and selectively processes the image features according to the mechanism of the gate unit. In this way, the LSTM network processes the image feature sequence frame by frame, and according to the processing result of each time step, classifies and predicts whether the input image feature sequence contains the state of the double flash signal lamp being turned on.

[0041] In step S2022, the parameters of the time sequence network model are updated based on the difference between the detection result and the label.

[0042] Specifically, the step S2022 includes:

[0043] In step a1, a loss function is determined based on the difference between the detection result and the label. Specifically, the loss function can adopt a cross-entropy loss function, and the loss function can be determined by the difference between the prediction result and the label.

[0044] In step a2, the parameters of the time sequence network model are updated by using the gradient descent method based on the loss function. Specifically, the loss function can be iteratively solved step by step by using the gradient descent method, and the model parameter value is obtained.

[0045] For example, the parameters of the time sequence network model F are solved by solving the following optimization problem: θ

[0046]

[0047] where the conditional probability CE(F​θ (x T ),c) describes the difference between the detection result and the true label. In short, the overall optimization goal is to minimize the difference between the recognition result of the time sequence network model and the true label. In the formula, x T represents each sample in the sample set X T .

[0048] Step S2023, the steps of repeating detection and parameter updating using the time sequence network model are repeated until the preset training round number is reached. Specifically, after updating the model parameters, the sample is re-input into the time sequence network model after parameter updating to obtain a new prediction result, the difference between the new prediction result and the label is used to determine the loss function, and is solved to realize model parameter updating; then the process is repeated until the preset training round number is reached, and the model at this time can be used as a vehicle double flash signal lamp recognition model.

[0049] As a specific application example of the embodiment of the application, as shown in the figure, the vehicle double flash signal lamp recognition model construction method is implemented by using the following flow: Figure 2

[0050] Step 1) First, collect a large amount of video data of vehicle tail double flash signal lamp opening or closing, divide the continuous frames at a certain time interval T to obtain the input sample x, and assign the label of double flash signal lamp opening or closing;

[0051] Step 2) determine the training round number and initialize the counter:

[0052] Specifically, the training round number n is determined, and the counter i is initialized (i=0).

[0053] Step 3) input sample x and update the counter:

[0054] Specifically, the sample x composed of continuous T frames of images is input, and the counter i is updated (i=i+1).

[0055] Step 4) the time sequence network model F outputs the detection result c':

[0056] Specifically, the sample x is input into the time sequence network model F, first through the CNN (convolutional neural network), the features of each frame of image are extracted, then the features are sent into the self-attention mechanism network for feature processing, and finally the accumulated T frame features are sent into the LSTM network to obtain the detection result c'.

[0057] Step 5) calculate the loss function using the detection result:

[0058] Specifically, the loss function is calculated using the detection result obtained in step 4).

[0059] ​Step 6) updating the parameters of the time sequence network model F by using gradient descent:

[0060] According to the loss function obtained in step 5), the parameters of the time sequence network model F are updated by using the gradient descent method.

[0061] Step 7) determining whether the counter is equal to the training round number:

[0062] Specifically, it is determined whether the current counter i is equal to the training round number n. If yes, the training is ended; otherwise, steps 2) to 6) are repeatedly executed.

[0063] Step 8) obtaining the time sequence classification model:

[0064] The trained time sequence classification model F, i.e., the vehicle double flash signal lamp recognition model, is obtained.

[0065] In this embodiment, a vehicle double flash signal lamp recognition method is provided, as shown in the figure, the method comprises the following steps: Figure 3

[0066] Step S301, acquiring continuous multiple frames of vehicle tail image data. Specifically, video data of a vehicle tail in a period of time to be recognized can be acquired, and then a fixed number of sliding windows are used to process the video data to obtain continuous multiple frames of image data.

[0067] Step S302, inputting the multiple frames of vehicle tail image data into the vehicle double flash signal lamp recognition model constructed by the vehicle double flash signal lamp recognition model construction method in the above embodiment to obtain the recognition result of whether the double flash signal lamp is turned on. Specifically, when the sliding window is used to generate image data, the image data in the window can be input into the model as the model input by moving one frame, and the recognition result of whether the double flash signal lamp is turned on in the image data is obtained through the processing of the model.

[0068] The vehicle double flash signal lamp recognition method provided in the embodiment of the application can obtain the recognition result of whether the double flash signal lamp is turned on by inputting continuous multiple frames of vehicle tail image data into the vehicle double flash signal lamp recognition model. Since the vehicle double flash signal lamp recognition model is trained by using the time sequence network model, and the state of the double flash signal lamp is flickering, the time sequence network can well capture the time sequence, and therefore, the recognition method can accurately recognize whether the double flash signal lamp is turned on.

[0069] ​A vehicle double flash signal lamp recognition model construction device is also provided in the embodiment, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described herein again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, implementation of hardware, or a combination of software and hardware, is also possible and contemplated.

[0070] The embodiment provides a vehicle double flash signal lamp recognition device, as shown in Figure 4 The device comprises:

[0071] A sample set construction module 401 is configured to obtain a sample set of continuous frame image data of a vehicle tail double flash signal lamp being turned on or off, and the image data is provided with a label of the double flash signal lamp being turned on or off.

[0072] A model construction module 402 is configured to train a time sequence network model by using the sample set, to obtain a vehicle double flash signal lamp recognition model, and the time sequence network model comprises a convolutional neural network, a self-attention mechanism network and a recurrent neural network.

[0073] Further function descriptions of the above-mentioned modules and units are the same as those of the corresponding embodiments, and will not be described herein again.

[0074] A vehicle double flash signal lamp recognition device is also provided in the embodiment, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described herein again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, implementation of hardware, or a combination of software and hardware, is also possible and contemplated.

[0075] The embodiment provides a vehicle double flash signal lamp recognition device, as shown in Figure 5 The device comprises:

[0076] A data acquisition module 501 is configured to acquire continuous multiple frames of vehicle tail image data.

[0077] An identification module 502 is configured to input the vehicle tail image data into a vehicle double flash signal lamp recognition model constructed by the vehicle double flash signal lamp recognition model construction method described in the above-mentioned embodiments, to obtain an identification result of whether the double flash signal lamp is turned on.

[0078] Further function descriptions of the above-mentioned modules and units are the same as those of the corresponding embodiments, and will not be described herein again.

[0079] The embodiment of the present application also provides a computer device with the above-mentioned Figure 4The vehicle double flash signal lamp recognition model construction device and Figure 5 The vehicle double flash signal lamp recognition device.

[0080] Please refer to Figure 6 , Figure 6 is a structural schematic diagram of a computer device provided by an optional embodiment of the present application, as Figure 6 indicated, the computer device includes one or more processors 10, a memory 20, and an interface for connecting components, including a high-speed interface and a low-speed interface. Various components are communicatively connected to each other by using different buses, and can be installed on a common mainboard or in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memories, if necessary. Similarly, multiple computer devices can be connected, each providing part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 In the above embodiment, the processor 10 is taken as an example.

[0081] The processor 10 can be a central processor, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic gate array, a generic array logic, or any combination thereof.

[0082] The memory 20 stores instructions executable by the at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0083] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function; the data storage area can store data created by use of the computer device according to the display of a small program landing page, and the like. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some optional embodiments, the memory 20 can optionally include a memory remotely arranged with respect to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0084] The memory 20 can include a volatile memory, such as a random access memory, and / or can include a non-volatile memory, such as a flash memory, a hard disk or a solid state disk. The memory 20 can also include a combination of the above-mentioned types of memories.

[0085] The computer device also comprises a communication interface 30 for enabling the computer device to communicate with other devices or communication networks.

[0086] The embodiments of the present application also provide a computer readable storage medium, the method according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as computer code recorded in a storage medium, or be implemented through network downloading and originally stored in a remote storage medium or a non-transitory machine readable storage medium and then stored in a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor or programmable or special purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that the computer, the processor, the microprocessor controller or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor or the hardware, the method shown in the above embodiments is implemented.

[0087] Although the embodiments of the present application have been described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for constructing a vehicle double flash signal lamp recognition model, characterized in that, The method comprises: obtaining continuous frame image data of a vehicle tail double flash signal lamp being turned on or off to construct a sample set, the image data being provided with a label of the double flash signal lamp being turned on or off; training a time sequence network model by using the sample set to obtain a vehicle double flash signal lamp recognition model, the time sequence network model comprising a convolutional neural network, a self-attention mechanism network and a recurrent neural network; the recurrent neural network is a long short-term memory network, and training the time sequence network model by using the sample set to obtain the vehicle double flash signal lamp recognition model comprises: inputting preset frame image data in the sample set into the time sequence network model, extracting features of each frame image by using the convolutional neural network, inputting the extracted features into the self-attention mechanism network to calculate similarities between the features, inputting features corresponding to multiple frames and having a similarity greater than a threshold value into the long short-term memory network, and outputting a detection result of whether the double flash signal lamp is turned on; updating parameters of the time sequence network model based on a difference between the detection result and the label; repeating the steps of detecting by using the time sequence network model and updating the parameters until a preset training round number is reached.

2. The method of claim 1, wherein, The method comprises: obtaining continuous frame image data of a vehicle tail double flash signal lamp being turned on or off to construct a sample set, the image data being provided with a label of the double flash signal lamp being turned on or off, obtaining video data when the vehicle tail double flash signal lamp is turned on and off; dividing the video data according to a preset time interval to obtain multiple continuous frame image data; 3. The method of claim 1, wherein, labeling each continuous frame image data as the double flash signal lamp being turned on or off to form the sample set. The method comprises: updating the parameters of the time sequence network model based on a difference between the detection result and the label comprises: 4.A vehicle double flash signal lamp recognition model construction device, characterized by determining a loss function based on the difference between the detection result and the label; updating the parameters of the time sequence network model by using a gradient descent method based on the loss function. The device comprises: a sample set construction module configured to obtain continuous frame image data of a vehicle tail double flash signal lamp being turned on or off to construct a sample set, the image data being provided with a label of the double flash signal lamp being turned on or off; a model construction module configured to train a time sequence network model by using the sample set to obtain a vehicle double flash signal lamp recognition model, the time sequence network model comprising a convolutional neural network, a self-attention mechanism network and a recurrent neural network; the recurrent neural network is a long short-term memory network, and training the time sequence network model by using the sample set to obtain the vehicle double flash signal lamp recognition model comprises: inputting preset frame image data in the sample set into the time sequence network model, extracting features of each frame image by using the convolutional neural network, inputting the extracted features into the self-attention mechanism network to calculate similarities between the features, inputting features corresponding to multiple frames and having a similarity greater than a threshold value into the long short-term memory network, and outputting a detection result of whether the double flash signal lamp is turned on; 5. A vehicle double flash signal lamp recognition method characterized by, updating parameters of the time sequence network model based on a difference between the detection result and the label; repeating the steps of detecting by using the time sequence network model and updating the parameters until a preset training round number is reached. The method comprises: obtaining continuous multiple frame vehicle tail image data; The multiple frames of vehicle tail image data are input into the vehicle double flashing signal lamp recognition model constructed by the vehicle double flashing signal lamp recognition model construction method in any one of claims 1-3 to obtain a recognition result of whether the double flashing signal lamp is on.

6. A vehicle double flasher signal recognition apparatus characterized by comprising: The device comprises: a data acquisition module configured to acquire continuous multiple frames of vehicle tail image data; an identification module configured to input the vehicle tail image data into the vehicle double flashing signal lamp recognition model constructed by the vehicle double flashing signal lamp recognition model construction method in any one of claims 1-3 to obtain a recognition result of whether the double flashing signal lamp is on.

7. A computer device, characterized by comprise: a memory and a processor, which are in communication connection with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the vehicle double flashing signal lamp recognition model construction method in any one of claims 1-3 or the vehicle double flashing signal lamp recognition method in claim 5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the vehicle double flashing signal lamp recognition model construction method in any one of claims 1-3 or the vehicle double flashing signal lamp recognition method in claim 5.

Images