TW202349266A - Method of recognizing moving vehicle and vehicle-mounted device and computer-readable storage medium - Google Patents

Abstract

The present application provides a method of recognizing a moving vehicle and vehicle-mounted device and computer-readable storage medium. The method is applied in a device carried on a vehicle. The method includes: capturing a first image of a target vehicle at a first moment and a second image of the target vehicle at a second moment; based on an instance segmentation algorithm, determining a first mask of the target vehicle in the first image and determining a second mask of the target vehicle in the second image; calculating an intersection over union (IoU) of the first mask and the second mask; determining whether an object mask in a dynamic class of the target vehicle is generated according to the IoU; in response that the object mask in the dynamic class of the target vehicle is generated, determining that the target vehicle is a moving vehicle. The present application can improve an accuracy of recognizing surrounding moving vehicles during driving a vehicle.

Description

Identification methods and related equipment for moving vehicles

The present invention relates to the technical field of target detection, and in particular to a method for identifying moving vehicles and related equipment.

When a vehicle is driving, it needs to make correct judgments about the surrounding environment and sense the dynamic objects around the vehicle to prevent traffic accidents. The objects in the image can be identified by performing image processing on the captured image. For example, semantic segmentation of image segmentation is used to identify and classify objects around the vehicle. For each primitive in the image, Assign a category, for example, vehicles and pedestrians are two categories respectively. However, the above methods cannot distinguish different objects in the same category (for example, vehicles), which may easily cause stationary vehicles to be recognized as moving vehicles at the same time when identifying moving vehicles, affecting the accuracy of identifying moving vehicles.

In view of the above, it is necessary to provide a moving vehicle identification method and related equipment, which solves the problem of low accuracy in identifying moving vehicles around the vehicle.

This application provides a method for identifying a moving vehicle, which is applied to an on-board device of the vehicle. The method includes: taking a first image of the target vehicle at a first moment and a second image at a second moment; based on an instance segmentation algorithm. , determine the first mask of the target vehicle according to the first image and determine the second mask of the target vehicle according to the second image; calculate the first mask and the second mask The intersection union ratio of the mask; determine whether to generate a dynamic class object mask of the target vehicle according to the intersection union ratio; if a dynamic class object mask of the target vehicle is generated, determine that the target vehicle is a moving vehicle dynamic Class object mask.

The method can calculate the intersection and union ratio of the corresponding masks of the target vehicle at different times to determine the dynamic object mask of the target vehicle, thereby improving the accuracy of identifying moving vehicles.

In some optional implementations, before determining the first mask of the target vehicle based on the first image, the method further includes: determining the location of the target vehicle in the first image. The first location and the identification of the target vehicle.

In some optional implementations, before determining the second mask of the target vehicle based on the second image, the method further includes: determining the target vehicle based on the identity of the target vehicle. at the second position in the second image.

In some optional implementations, the instance-based segmentation algorithm determines a first mask of the target vehicle based on the first image and determines a second mask of the target vehicle based on the second image. Masking includes: inputting the first image and the second image into a feature extraction network respectively, and obtaining a first feature map of the first image and a second feature map of the second image. ; Perform binary classification and coordinate regression on the first feature map and the second feature map respectively, and determine the first area of interest of the target vehicle in the first feature map and the first area of interest in the second feature map. The second area of interest of the target vehicle; extracting the first feature sub-image of the first area of interest and the second feature sub-image of the second area of interest; based on the first feature sub-image and the The first mask is generated at a first location; the second mask is generated based on the second feature submap and the second location.

In some optional implementations, calculating the intersection union ratio of the first mask and the second mask includes: calculating the first intersection ratio of the first mask and the second mask. The degree of coincidence; calculating the second degree of coincidence between the first mask and the second mask; calculating the ratio of the first degree of coincidence to the second degree of coincidence as the intersection union ratio.

In some optional implementations, determining whether to generate a dynamic class object mask for the target vehicle based on the intersection-to-union ratio includes: if the intersection-to-union ratio is greater than or equal to a preset threshold, determining to generate The dynamic object mask of the target vehicle; if the intersection and union ratio is less than the preset threshold, it is determined not to generate the dynamic object mask of the target vehicle.

In some optional implementations, after determining that the target vehicle is a moving vehicle, the method further includes: establishing a vehicle safe distance model, and obtaining a safe braking distance between vehicles based on the vehicle safe distance model; if it is determined that the If the distance between the moving vehicle and the vehicle is less than or equal to the safe braking distance, an alarm is issued.

In some optional implementations, the alarm includes a first alarm and a second alarm, and issuing the alarm includes: if the distance between the moving vehicle and the vehicle is equal to the safe braking distance, issuing the alarm at a first frequency. The first alarm; if the distance between the moving vehicle and the vehicle is less than the safe braking distance, the second alarm is issued at a second frequency; wherein the first frequency is less than the second frequency.

This application also provides a vehicle-mounted device. The vehicle-mounted device includes a processor and a storage. The processor is configured to implement the mobile vehicle identification method when executing a computer program stored in the storage.

This application also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the mobile vehicle identification method is implemented.

Compared with the prior art, the mobile vehicle identification method and related equipment provided by this application can accurately identify each vehicle among multiple vehicles in the same scene, regard each vehicle as a target vehicle, and calculate each target vehicle. The intersection and union ratios of the corresponding masks of the vehicle at different times are used to determine whether the target vehicle is moving based on the intersection and union ratio, which improves the accuracy of identifying moving vehicles and thereby improves the safety of vehicle driving.

To facilitate understanding, some descriptions of concepts related to the embodiments of the present application are exemplarily provided for reference.

It should be noted that “at least one” in this application refers to one or more, and “multiple” refers to two or more than two. "And/or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and/or B can mean: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can Is singular or plural. The terms "first", "second", "third", "fourth", etc. (if present) in the description, claim and drawings of this application are used to distinguish similar objects, rather than to Describe a specific order or sequence.

In order to better understand the mobile vehicle identification method and related equipment provided by the embodiments of the present application, the application scenarios of the mobile vehicle identification method of the present application are first described below.

Figure 1 is a schematic structural diagram of a vehicle-mounted device 1 provided by an embodiment of the present application. Referring to FIG. 1 , the vehicle-mounted device 1 includes, but is not limited to, a memory 11 and at least one processor 12 . The storage 11 and the processor 12 can be connected through the communication bus 13 or directly. The vehicle-mounted device 1 is disposed on a vehicle. The vehicle-mounted device 1 may be a vehicle-mounted computer. In some embodiments, the vehicle-mounted device 1 may include a shooting device to capture multiple images or videos around the vehicle. In other implementations, For example, the vehicle-mounted device 1 may not include a shooting device, but may establish a communication bus connection with one or more shooting devices inside the vehicle to directly obtain multiple images or videos from the shooting devices of the vehicle, for example , you can also connect to the communication bus with the driving recorder in the vehicle and obtain the corresponding images or videos.

Those skilled in the art should understand that the structure of the vehicle-mounted device 1 shown in Figure 1 does not constitute a limitation of the embodiment of the present invention. The vehicle-mounted device 1 may also include more or less other hardware or software than in Figure 1 . Or different widget settings.

The processor 12 in the vehicle-mounted device 1 can implement a moving vehicle identification method that will be described in detail below when executing a computer program. The computer program includes a moving vehicle identification program.

When a vehicle is driving, it needs to make correct judgments about the surrounding environment and sense the dynamic objects around the vehicle to prevent traffic accidents. The captured images can be processed through image processing to identify objects in the image. For example, semantic segmentation of image segmentation is used to identify and classify objects around the vehicle, and each primitive in the image is assigned A category, for example, vehicles and pedestrians are two categories respectively. However, the above method cannot distinguish different objects in the same category (for example, vehicles), which may easily cause stationary vehicles to be recognized as moving vehicles at the same time when identifying moving vehicles, affecting the accuracy of identification.

In order to solve the above technical problems, an embodiment of the present application provides a method for identifying a moving vehicle. As shown in Figure 2, it is a flow chart of the method for identifying a moving vehicle provided by an embodiment of the present application. The identification method of moving vehicles is applied in the vehicle-mounted device 1, which can improve the accuracy of identifying moving vehicles and ensure the safety of vehicle driving. Depending on different needs, the order of steps in this flowchart can be changed and some steps can be omitted. In this embodiment, the identification method of a moving vehicle includes the following steps:

21. Capture a first image of the target vehicle at a first moment and a second image at a second moment.

In embodiments of the present application, multiple images of the surroundings of the vehicle at different times can be captured using a vehicle's photography device (for example, a monocular camera). Specifically, images containing multiple target objects can be captured, and the target objects can be objects of different categories, such as target vehicles or people, and two images captured at consecutive times are used as the first image at the first moment and the A second image at a second moment. Taking the target vehicle as an example, a first image of the target vehicle at a first time and a second image at a second time are captured.

In the image of the scene outside the vehicle captured by the shooting device, the captured RGB image is used as a single image. The embodiment of the present application obtains the first image at the first time and the second image at the second time respectively. picture. Suppose we obtain two consecutive images in a scene , then Represents the first image at time t (i.e. the first moment), Represents the second image at time t+1 (ie, the second time).

22. Based on the instance segmentation algorithm, determine the first mask of the target vehicle according to the first image and determine the second mask of the target vehicle according to the second image.

The instance segmentation algorithm (Instance Segmentation) has the characteristics of both semantic segmentation and target detection. That is, the instance segmentation algorithm can classify at the primitive level and locate different instances. For example, both It can distinguish multiple objects of the same category as a single target object, and can also classify the target object into a vehicle and locate the vehicle location. In the embodiment of the present application, multiple identified vehicles are distinguished and each vehicle is identified. The vehicle is used as a separate instance object, and the instance object is the target vehicle.

Mask, also known as masking, means using a selected image, graphic or object to block the image to be processed (all or part of it), thereby controlling the area or process of image processing. For example, if an object in an image to be processed is blocked, the blocked area is called a mask. In the embodiment of the present application, each identified target vehicle is processed using different masks to distinguish different target vehicles in the same image.

In one embodiment of the present application, the first image and the second image are respectively input into the feature extraction network to obtain the feature map. Specifically, the first image is input into the feature extraction network to obtain the third image of the first image. A feature map, inputting the second image into the feature extraction network to obtain the second feature map of the second image. The first image at time t and the second image at time t+1 are respectively used as inputs of the monocular depth estimation residual convolutional neural network model, where the monocular depth estimation residual convolutional neural network model includes One input layer, seven convolution layers, seven deconvolution layers and four residual terms are calculated by using the monocular depth estimation residual convolutional neural network model to obtain the corresponding convolution of the first image at time t The first feature map, and the second feature map corresponding to the second image at time t+1 after convolution.

In one embodiment of the present application, binary classification and coordinate regression are performed on the first feature map and the second feature map respectively to determine whether the first feature map contains a first region of interest (Region of interest) of the target vehicle. Interest, ROI) and determine whether the second feature map contains a second region of interest (Region of Interest, ROI) of the target vehicle.

Specifically, after the convolution layer outputs the first feature map, a depth linear regression function is added to map each primitive point in the first feature map to the corresponding depth value. Then set a fixed number (default 15) of first regions of interest for each primitive position in the first feature map, and input these first regions of interest into the Region Proposal Network (RPN). Perform binary classification and coordinate regression to obtain the first area of interest where the target vehicle exists, and use the same method to obtain the second area of interest of the second feature map. After obtaining the first region of interest and the second region of interest in which the target vehicle exists, the first feature subgraph is extracted from each first region of interest through the regional feature aggregation method (ROIAlign), and these first features are The sub-pictures are classified, and candidate frame regression is performed on the first feature sub-picture after classification and a fully convolutional network (Fully Convolutional Network, FCN) is introduced to generate the first mask. Similarly, ROIAlign is used to extract the second feature sub-image from each second region of interest, classify these second feature sub-images, and perform candidate frame regression and introduce full convolution on the classified second feature sub-image. The network (Fully Convolutional Network, FCN) generates the second mask.

Region of interest pooling (ROI-Pooling) can usually be used to convert regions of interest (ROI) of different sizes into fixed-size ROIs. However, during the conversion process, two rounds of rounding and quantification are required, resulting in the feature space being transformed. There will be a large deviation when the ROI corresponds to the original image, affecting the accuracy of image segmentation. The embodiment of the present application solves the problem of inaccurate regions after segmentation by using ROIAlign proposed on the basis of ROI-Pooling. ROIAlign cancels the rounding operation, retains all floating point numbers, and then obtains the values of multiple sampling points through bilinear interpolation, and then performs maximum value pooling on multiple sampling points to obtain the Click the final value to improve the accuracy of the detection model.

In one embodiment of the present application, before generating the first mask, an instance segmentation algorithm may also be used to obtain the first position of the target vehicle in the first image and the identification of the target vehicle. The identification is Markings for each vehicle. According to the identification of the target vehicle obtained at the first moment, when the second image is obtained at the second moment, the second position of the target vehicle in the second image is located, and then through instance segmentation The algorithm performs segmentation processing on the target vehicle in the second image to generate a second mask at the second position.

23. Calculate the intersection union ratio of the first mask and the second mask.

A first mask of a first position of the target vehicle in the first image and a second mask of a second position of the target vehicle in the second image are obtained based on 22 . The calculation method of Intersection over Union (IoU) is used to describe the coincidence degree of the first mask at the first position and the second mask at the second position in the same scene, including calculating the sum of the first mask and The first degree of coincidence of the second mask, and the second degree of coincidence of the first mask and the second mask. For example, for a target vehicle in the same scene, the positional movement degree of the target vehicle is calculated based on the first mask of the target vehicle at the first position at time t and the second mask of the target vehicle at the second position at time t+1.

Please refer to Figure 3, which is a flow chart for calculating the intersection and union ratio of the target vehicle at different times.

The degree of overlap between the first mask and the second mask includes the intersection of the first mask and the second mask (the first degree of overlap) and the union of the first mask and the second mask (the second degree of overlap).

31. Calculate the intersection of the first mask and the second mask.

The first mask at the first position and the second mask at the second position are respectively regarded as the area of a collection of primitives, and the area of the collection of primitives represented by the first mask and the area represented by the second mask are calculated. The intersection of regions of collections of primitives.

32. Calculate the union of the first mask and the second mask.

The first mask at the first position and the second mask at the second position are respectively regarded as the area of a collection of primitives, and the area of the collection of primitives represented by the first mask and the area represented by the second mask are calculated. The union of regions of collections of primitives.

33. Determine the intersection-to-union ratio based on the ratio of the intersection to the union.

In one embodiment of the present application, the ratio of the intersection of the first mask and the second mask to the union of the first mask and the second mask is used as the intersection-to-union ratio.

The first mask in the first image and the second mask in the second image are respectively regarded as the area of a primitive set, and the primitive collection area obtained at time t is regarded as the first area, and t+1 The collection area of primitives obtained at any time is used as the second area, and the intersection union ratio of the first area and the second area is calculated. Specifically, the area of the overlap (intersection) of the first area and the second area is divided by the area of the first area and the second area. The area of the combined portion (union) of the region.

For example, assume that the positions of the first region and the second region in the coordinate system are respectively expressed as [ ], [ ],in,( ) represents the coordinates of the upper left corner of the first area, ( ) represents the coordinates of the lower right corner of the first area, ( ) represents the coordinates of the upper left corner of the second area, ( ) represents the coordinates of the lower right corner of the second area, then the coordinates of the upper left corner of the intersection area between the first area and the second area are calculated as: , , the coordinates of the lower right corner of the intersection area are: , .

;

Calculate the area S11 of the first region and the area S22 of the second region respectively, as follows:

;

;

Calculate the area S2 of the union of the first area area and the second area area, as follows:

S2=S11+S22-S1;

According to the above area S1 of the intersection part and the area S2 of the union part, calculate the intersection union ratio of the first region and the second region, as follows:

IoU=S1/S2.

24. Determine whether to generate a dynamic class object mask of the target vehicle according to the intersection union ratio.

The dynamic object mask refers to a mask formed by connecting the masks of the target vehicle at different times. For example, a moving vehicle forms a first mask when it is in the first position at time t, and forms a second mask when it moves to the second position at time t+1. The first mask and the second mask are connected. The mask formed serves as a marker indicating that the vehicle is moving, i.e., a dynamic class object mask.

Determine whether to generate the dynamic object mask of the target vehicle based on the intersection union ratio obtained in step 23. Before determining whether to generate the dynamic object mask of the target vehicle, determine the state of the target vehicle, and the state includes Stationary state and moving state. If it is determined that the state of the target vehicle is in a moving state, then the first mask and the second mask are connected to generate a dynamic object mask of the target vehicle. If it is determined that the state of the target vehicle is in a stationary state, no generation will be performed. Dynamic class object mask of the target vehicle.

Please refer to Figure 4, which is a flow chart for determining whether to generate a dynamic class object mask of the target vehicle.

41. Compare the intersection union ratio of the first mask and the second mask with the preset threshold.

The intersection union ratio of the first mask and the second mask is between 0 and 1. If the coincidence degree of the first mask and the second mask is close to 1, then the coincidence degree of the first mask and the second mask is high. If the coincidence degree of the first mask and the second mask is close to 0, then the coincidence degree of the first mask and the second mask is close to 0. The overlap between the mask and the second mask is low. Taking the target vehicle as an example, if the overlap between the first mask of a certain target vehicle at time t and the second mask at time t+1 is close to 1, then the first mask of the target vehicle at time t is consistent with the first mask at time t. The second mask at time t+1 has a high degree of overlap. If the overlap between the first mask of the target vehicle at time t and the second mask at time t+1 is close to 0, then the first mask of the target vehicle at time t and the second mask at time t+1 Mask overlap is low. In one embodiment of the present application, the preset threshold can be set at 0.5, and a certain error can be allowed. For example, the above comparison can be completed by using a preset error range combined with the preset threshold.

42. If the intersection union ratio of the first mask and the second mask is greater than or equal to the preset threshold, determine to generate a dynamic class object mask of the target vehicle.

For example, if the coincidence degree between the first mask of a certain target vehicle at time t and the second mask at time t+1 is greater than or equal to 0.5, a dynamic class object mask of the target vehicle is generated.

43. If the intersection union ratio of the first mask and the second mask is less than the preset threshold, determine not to generate a dynamic object mask of the target vehicle.

For example, if the coincidence degree between the first mask of a certain target vehicle at time t and the second mask at time t+1 is less than 0.5, no dynamic class object mask of the target vehicle is generated.

25. If the dynamic class object mask of the target vehicle is generated, determine that the target vehicle is a moving vehicle.

If the coincidence degree between the first mask of a certain target vehicle at time t and the second mask at time t+1 is greater than or equal to 0.5, a dynamic class object mask of the target vehicle is generated, that is, the target vehicle is connected The mask formed by the masks at different times is used as the dynamic class object mask of the target vehicle, and the vehicle that generates the dynamic class object mask is determined to be a moving vehicle.

Usually, the scene is fixed. If there are moving objects, such as cars and pedestrians, during the shooting process of the monocular camera, the calculation of the loss function will be inaccurate. Because the car is moving, the loss function will be very large. For example, you can use the method of semantic segmentation to mask out all moving objects when calculating the loss function. Since semantic segmentation can only divide different types of objects, therefore, in the process of masking out all moving objects, , it is easy to mask the stopped objects together, affecting the accuracy of recognition.

When a vehicle is driving, it is necessary to make accurate judgments on the status of surrounding vehicles and pedestrians. In the process of identifying objects, the embodiment of the present application uses instance segmentation technology to segment the bullseye chart image, and accurately identifies each object. , for example, when multiple vehicles are identified, how many vehicles can be identified, and moving vehicles and stationary vehicles can be identified.

In one embodiment of the present application, after determining that the target vehicle is a moving vehicle, a vehicle safety distance model is established, and a safe braking distance between vehicles is obtained based on the vehicle safety distance model. If it is determined that the distance between the moving vehicle and the vehicle is less than Or equal to the safe braking distance, an alarm will sound. The alarm includes a first alarm and a second alarm. If the distance between the moving vehicle and the vehicle is equal to the safe braking distance, the first alarm is issued at a first frequency; if the distance between the moving vehicle and the vehicle is the distance is less than the safe braking distance, the second alarm is issued at a second frequency; wherein the first frequency is less than the second frequency.

In practical applications, the purpose of calculating the mask of a moving vehicle is to more accurately determine the distance between the own vehicle and surrounding vehicles/pedestrians to ensure safe driving.

Specifically, by determining that the surrounding target vehicles are moving vehicles, a vehicle safety distance model between the surrounding moving vehicles and the own vehicle is established.

A vehicle safety distance model is constructed based on the calculation of the speed v1 of the vehicle at time t, the speed v2 of the surrounding vehicles (i.e., the target vehicle), and the distance warning distance d between the vehicle and the surrounding vehicles. Based on the vehicle safety distance model, the safe braking distance between vehicles is obtained. .

If the distance between the two vehicles is greater than the safe braking distance, the vehicle can drive normally and no alarm is required. If the distance between the two vehicles is less than or equal to the safe braking distance, an alarm will sound to remind the driver that he is too close to the surrounding vehicles. The alarm includes a first alarm and a second alarm. If the distance between the moving vehicle and the vehicle is equal to the safe braking distance, the first alarm is issued at a first frequency. If the distance between the moving vehicle and the vehicle is less than the safe braking distance, the first alarm is issued at the second frequency. For the second alarm, the first frequency is less than the second frequency. For example, when it is determined that the distance between the moving vehicle and the vehicle is equal to the safe braking distance, a first alarm is issued at a frequency of 5 times per minute, and when the distance between the moving vehicle and the vehicle is less than the safe braking distance, a second alarm is issued at a frequency of 10 times per minute. , and when the time for issuing the second alarm exceeds the preset time, for example, when the driver does not make any operation to avoid danger within 2 minutes, the automatic emergency braking system is activated to ensure the safety of the vehicle.

In this embodiment, the safe braking distance d can also be divided into three different critical distances according to different safety warning forms: the critical safety distance, the critical dangerous distance and the critical infinitesimal distance. . Different safety warning measures are taken based on the distance between the vehicle and the vehicle.

This application uses an instance segmentation algorithm to improve the accuracy of identifying surrounding moving vehicles, and further ensures the monitoring of vehicle safety distances while the vehicle is driving.

Please continue to refer to FIG. 1 . In this embodiment, the memory 11 may be an internal memory of the vehicle-mounted device 1 , that is, a memory built into the vehicle-mounted device 1 . In other embodiments, the storage 11 may also be an external storage of the vehicle-mounted device 1 , that is, a storage external to the vehicle-mounted device 1 .

In some embodiments, the memory 11 is used to store program codes and various data, and realize high-speed and automatic access to programs or data during the operation of the vehicle-mounted device 1 .

The storage 11 may include random access memory, and may also include non-volatile storage, such as a hard disk, a storage device, a plug-in hard disk, a Smart Media Card (SMC), a secure digital ( Secure Digital (SD) card, flash memory card (Flash Card), at least one disk storage device, flash memory device, or other volatile solid-state storage device.

In one embodiment, the processor 12 may be a central processing unit (CPU), or other general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), or an application specific integrated circuit (Application Processor). Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any other conventional processor, etc.

If the program codes and various data in the storage 11 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above embodiment methods, such as the identification method of moving vehicles, which can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in the computer. In the readable storage medium, when the computer program is executed by the processor, the steps of the above method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in the form of original program code, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer storage, and a read-only memory (ROM, Read- Only Memory) etc.

It can be understood that the module division described above is a logical function division, and there may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application can be integrated in the same processing unit, or each module can exist physically alone, or two or more modules can be integrated in the same unit. The above integrated modules can be implemented in the form of hardware or in the form of hardware plus software function modules.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified. Modifications or equivalent substitutions may be made without departing from the spirit and scope of the technical solution of the present application.

1: Vehicle-mounted device 11:Storage 12: Processor 13: Communication bus 21~25: Steps 31~33: Steps 41~43: Steps

Figure 1 is a schematic structural diagram of a vehicle-mounted device according to a preferred embodiment of the present invention.

Figure 2 is a flow chart of a moving vehicle identification method provided by a preferred embodiment of the present invention.

Figure 3 is a flow chart for calculating the intersection union ratio of the target vehicle at different times.

Figure 4 is a flow chart for determining whether to generate a dynamic class object mask of the target vehicle.

21~25: Steps

Claims (10)

A method for identifying a moving vehicle, applied to an on-board device of the vehicle, wherein the method includes: Capture a first image of the target vehicle at a first moment and a second image at a second moment; Based on an instance segmentation algorithm, determine a first mask of the target vehicle based on the first image and determine a second mask of the target vehicle based on the second image; Calculate the intersection union ratio of the first mask and the second mask; Determine whether to generate a dynamic class object mask of the target vehicle according to the intersection union ratio; If a dynamic object mask of the target vehicle is generated, it is determined that the target vehicle is a moving vehicle. The method for identifying a moving vehicle as claimed in claim 1, wherein before determining the first mask of the target vehicle based on the first image, the method further includes: A first position of the target vehicle in the first image and an identity of the target vehicle are determined. The method for identifying a moving vehicle according to claim 2, wherein before determining the second mask of the target vehicle based on the second image, the method further includes: Based on the identification of the target vehicle, a second location of the target vehicle in the second image is determined. The method for identifying a moving vehicle according to claim 3, wherein the instance-based segmentation algorithm determines the first mask of the target vehicle based on the first image and determines the target vehicle based on the second image. The second mask of the target vehicle includes: Input the first image and the second image into a feature extraction network respectively, and obtain the first feature map of the first image and the second feature map of the second image; Perform binary classification and coordinate regression on the first feature map and the second feature map respectively to determine the first area of interest of the target vehicle in the first feature map and the first area of interest in the second feature map. a second region of interest for the target vehicle; extracting a first feature sub-image of the first region of interest and a second feature sub-image of the second region of interest; generating the first mask based on the first feature submap and the first location; The second mask is generated based on the second feature submap and the second location. The identification method of a moving vehicle according to any one of claims 1 to 4, wherein the calculation of the intersection and union ratio of the first mask and the second mask includes: Calculate a first degree of coincidence between the first mask and the second mask; Calculate a second degree of coincidence between the first mask and the second mask; The ratio of the first degree of coincidence to the second degree of coincidence is calculated as the intersection union ratio. The identification method of a moving vehicle as described in claim 1, wherein the determining whether to generate a dynamic class object mask of the target vehicle based on the intersection union ratio includes: if the intersection union ratio is greater than or equal to A preset threshold is used to determine to generate a dynamic object mask for the target vehicle; if the intersection and union ratio is less than the preset threshold, it is determined not to generate a dynamic object mask for the target vehicle. The method for identifying a moving vehicle as claimed in claim 1, wherein after determining that the target vehicle is a moving vehicle, the method further includes: Establish a vehicle safety distance model, and obtain the safe braking distance between vehicles based on the vehicle safety distance model; If it is determined that the distance between the moving vehicle and the vehicle is less than or equal to the safe braking distance, an alarm is issued. The method for identifying a moving vehicle according to claim 7, wherein the alarm includes a first alarm and a second alarm, and the issuing of the alarm includes: If the distance between the moving vehicle and the vehicle is equal to the safe braking distance, send the first alarm at the first frequency; If the distance between the moving vehicle and the vehicle is less than the safe braking distance, send the second alarm at a second frequency; Wherein, the first frequency is smaller than the second frequency. A vehicle-mounted device, wherein the vehicle-mounted device includes a processor and a storage, and the processor is used to execute a computer program stored in the storage to implement the identification method of a moving vehicle according to any one of claims 1 to 8. . A computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction. When the at least one instruction is executed by a processor, the identification of a moving vehicle as described in any one of claims 1 to 8 is achieved. method.