A method and apparatus for recognizing a target object

Abstract

The application discloses a target object recognition method and device, and relates to the technical field of super deep learning. The specific implementation of the method comprises the following steps: acquiring current priorities of a plurality of target objects to be detected included in a vehicle terminal image; wherein the current priorities are determined according to driving data of the vehicle; determining a plurality of preprocessing regions of the vehicle terminal image according to target input scales of a pre-trained image detection model and the current priorities; wherein the preprocessing regions include one or more target objects to be detected; and processing the vehicle terminal image according to the plurality of preprocessing regions, taking a processing result as an input of the image detection model, and recognizing the target objects. The implementation can cover a larger number of target objects with higher priorities in a limited detection region without consuming additional vehicle terminal resources, thereby realizing multi-region detection of the target objects, improving user experience, and guaranteeing driving safety and stability of vehicle terminal services.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a target object recognition method and device. BACKGROUND

[0002] Image identification refers to processing, analyzing and understanding images using deep learning algorithms to identify various different patterns of targets and objects.

[0003] In the existing vehicle driving process, in order to identify target objects (such as traffic lights) in the field of view, the vehicle server usually determines a region of interest, adjusts the region of interest to the required scale for model detection through a resize function, and then performs model detection.

[0004] However, in the limited detection area of the model, if multiple target objects are included in the field of view, the existing cropping method cannot be balanced, which may cause target loss; if the batch size is increased for detection, the vehicle end resources required will be larger, causing the vehicle end server to have excessive load pressure, response delay, poor user experience, and in severe cases, even causing the vehicle end server to crash and causing safety accidents. SUMMARY

[0005] Therefore, the embodiments of the present application provide a target object recognition method and device, which can cover a larger number and higher priority of target objects in a limited detection area without consuming additional vehicle end resources, thereby realizing multi-region detection of target objects, improving user experience, and ensuring driving safety and stability of the vehicle end service.

[0006] To achieve the above object, according to an aspect of the embodiments of the present application, a target object recognition method is provided, comprising:

[0007] Obtaining the current priority of a plurality of target objects to be detected included in a vehicle end image; wherein the current priority is determined according to driving data of the vehicle;

[0008] Determining a plurality of preprocessing regions of the vehicle end image according to the target input scale of a pre-trained image detection model and the current priority; wherein the preprocessing region includes one or more target objects to be detected;

[0009] Processing the vehicle end image according to a plurality of preprocessing regions, taking the processing result as the input of the image detection model to identify the target object.

[0010] Optionally, the determining, according to the target input scale of the image detection model and the current priority, of the plurality of preprocessing regions of the vehicle-end image comprises:

[0011] determining, according to the target input scale, of a target region scale of the target object to be detected;

[0012] determining, according to a preset edge scale and the target region scale, of a candidate region set of a plurality of the target object to be detected; wherein the candidate region set comprises a plurality of candidate region frames, and the candidate region frames each include one or more target objects to be detected;

[0013] determining, according to a current priority and a target number of the target object to be detected in the candidate region set, of the plurality of preprocessing regions.

[0014] Optionally, the determining, according to a preset edge scale and the target region scale, of a candidate region set of a plurality of the target object to be detected comprises:

[0015] generating, according to the preset edge scale and the target region scale, a plurality of candidate region frames corresponding to the target object to be detected;

[0016] sorting the plurality of candidate region frames according to the scale of the candidate region frames;

[0017] determining, according to the sorting result, of a plurality of candidate regions of all the target objects to be detected included in the vehicle-end image, to form the candidate region set.

[0018] Optionally, the determining of a plurality of candidate regions of all the target objects to be detected included in the vehicle-end image comprises:

[0019] starting from a smallest candidate region frame, and performing the following steps in a loop until the candidate region includes a maximum number of target objects to be detected:

[0020] A1: determining, according to the scale of the candidate region frame, of one or more other target objects to be detected closest to the target object to be detected corresponding to the candidate region frame;

[0021] A2: determining whether a region scale including the target object to be detected corresponding to the candidate region frame and the other target objects to be detected is less than or equal to the scale of the candidate region frame, and if so, proceeding to step A3; if not, proceeding to step A4;

[0022] A3: adding the other target objects to be detected as the target object to be detected corresponding to the candidate region frame, and proceeding to step A1;

[0023] A4: taking the candidate region frame at any angle as a candidate region.

[0024] Optionally, the determining the plurality of preprocessing regions according to the current priority and the target number of the target object included in the candidate region set comprises:

[0025] determining a candidate priority of each of the candidate regions included in the candidate region set according to the current priority and the target number of the target object included in the candidate region;

[0026] determining the candidate region with the highest candidate priority as a first type of preprocessing region, and determining the remaining candidate regions as a second type of preprocessing region.

[0027] Optionally, the determining the candidate priority of each of the candidate regions included in the candidate region set according to the current priority and the target number of the target object included in the candidate region comprises:

[0028] in a case where the target number of the target object included in the candidate region is one, the candidate priority is the current priority of the target object;

[0029] in a case where the target number of the target object included in the candidate region is multiple, the candidate priority is a sum of the current priorities of the multiple target objects.

[0030] Optionally, the vehicle-side image corresponds to a camera coordinate system of the vehicle; and the determining the target region scale of the target object according to the target input scale comprises:

[0031] determining a real object size of the target object according to driving data of the vehicle;

[0032] determining the target region scale of the target object according to the real object size and the target input scale in the camera coordinate system.

[0033] Optionally, the processing the vehicle-side image according to the plurality of preprocessing regions and taking the processing result as the input of the image detection model comprises:

[0034] cropping the vehicle-side image according to the plurality of preprocessing regions and a preset cropping scale to obtain a plurality of cropped regions;

[0035] splicing the plurality of cropped regions;

[0036] determining the input of the image detection model according to the splicing result.

[0037] Optionally, the method further comprises:

[0038] According to the current detection result of the image detection model, the priority of the target object corresponding to the second type of pre-processing region is adjusted;

[0039] According to the adjusted priority of the target object, the current priority of the target object is determined;

[0040] According to the adjusted current priority of the target object, the target detection result of the image detection model is re-executed according to the target input scale of the image detection model and the current priority.

[0041] Optionally, further comprising:

[0042] The target detection result of the image detection model is normalized;

[0043] According to the processing result of the normalization processing, the output of the image detection model is determined;

[0044] According to the output of the image detection model, the target object is identified.

[0045] Optionally, the image detection model is trained according to the modified input scale; wherein the modified input scale corresponds to multi-region detection of multiple target objects.

[0046] According to another aspect of the embodiment of the application, a target object recognition device is provided, comprising:

[0047] The acquisition module is configured to acquire the current priority of the multiple inherent targets to be detected included in the vehicle-side image; wherein the current priority is determined according to the driving data of the vehicle;

[0048] The pre-processing module is configured to determine the multiple pre-processing regions of the vehicle-side image according to the target input scale of the pre-trained image detection model and the current priority; wherein the pre-processing region includes one or more of the inherent targets to be detected;

[0049] The recognition module is configured to process the vehicle-side image according to the multiple pre-processing regions, and take the processing result as the input of the image detection model to identify the inherent target.

[0050] According to another aspect of the embodiment of the application, an electronic device for recognizing a target object is provided, comprising:

[0051] One or more processors;

[0052] The storage device is configured to store one or more programs,

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the target object identification method provided by the present application.

[0054] According to still another aspect of the embodiments of the present application, a computer readable medium is provided, which has stored thereon a computer program, and the program is executed by a processor to implement the target object identification method provided by the present application.

[0055] One of the embodiments of the above-mentioned application has the following advantages or beneficial effects: because the existing image detection model is improved, the image detection model can perform multi-region detection, and then the target input scale of the trained image detection model is used to determine the target scale of the target object to be detected; the target region scale of the target object to be detected is determined according to the target scale; on the basis of the target region scale, the candidate region frame is constructed, so as to determine the candidate region set from the smallest candidate region frame, and to include as many target objects to be detected as possible; then, the candidate region priority is determined according to the initial priority of the target object to be detected and the target quantity, so as to obtain various types of pre-processing regions (including stable priority regions and dynamic switching regions); the pre-processing regions are cropped and spliced to generate the input of the image detection model; the priority of the target object is adjusted according to the detection result, so as to obtain the final target detection result; the target detection result is normalized and fused by NMS, so as to obtain the output of the image detection model, so as to determine the multi-region detection result of the vehicle-end image including multiple target objects to be detected. Therefore, the technical means of the above-mentioned application can overcome the technical problems that the target may be lost when the existing detection model detects the vehicle-end image of multiple target objects; by increasing the batch size, a large amount of vehicle-end resources needs to be consumed, so that the load pressure of the vehicle-end server is too large, the response delay is too long, the user experience is poor, and in serious cases, the vehicle-end server may even crash to cause a safety accident. Therefore, the technical effects of the above-mentioned application are achieved, that is, more number and higher priority target objects can be covered in the limited detection region without consuming additional vehicle-end resources, so that the multi-region detection of the target object is realized, the user experience is improved, and the stability of the driving safety and the vehicle-end service is ensured.

[0056] The further effects of the above-mentioned non-conventional optional mode will be described in combination with the specific embodiments below. BRIEF DESCRIPTION OF DRAWINGS

[0057] The accompanying drawings are used to better understand the present application, and do not constitute an improper limitation on the present application. Among them:

[0058] Figure 1 is a schematic diagram of the main process of the target object identification method according to the embodiments of the present application;

[0059] Figure 2is a schematic diagram of the main flow of a training method of an image detection model according to an embodiment of the present application;

[0060] Figure 3 is a schematic diagram of the main flow of a determination method of a pre-processing region according to an embodiment of the present application;

[0061] Figure 4 is a schematic diagram of the main flow of a determination method of a candidate region according to an embodiment of the present application;

[0062] Figure 5 is a schematic diagram of the main flow of a determination method of an input of an image detection model according to an embodiment of the present application;

[0063] Figure 6 is a schematic diagram of multi-region detection of an image detection model according to an embodiment of the present application;

[0064] Figure 7 is a schematic diagram of the main flow of an adjustment method of a priority of a target object according to an embodiment of the present application;

[0065] Figure 8 is a schematic diagram of the main flow of a determination method of an output of an image detection model according to an embodiment of the present application;

[0066] Figure 9 is a schematic diagram of the main modules of a target object recognition apparatus according to an embodiment of the present application;

[0067] Figure 10 An exemplary system architecture diagram suitable for applying a target object recognition method or a target object recognition apparatus according to an embodiment of the present application is shown.

[0068] Figure 11 is a structural schematic diagram of a computer system of a terminal device or a server suitable for implementing an embodiment of the present application. DETAILED DESCRIPTION

[0069] Exemplary embodiments of the present application are described herein with reference to the accompanying drawings, which are meant to be exemplary and not limiting. Therefore, it should be recognized that many modifications and changes can be made to the embodiments described herein without departing from the spirit and scope of the present application. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0070] Scale: i.e. scale, used to represent the pixel size of each region, such as a candidate region, a detection region, etc.; wherein the scale can be an interval value.

[0071] padding: the padding attribute defines the space between the element border and the element content. By setting the inner margin attribute of the element, such as increasing the number of pixels of each side, it is ensured that the feature map is not too small to prevent the loss of original image information.

[0072] Figure 1 is a schematic diagram of the main process of the target object recognition method according to an embodiment of the present application, as shown in Figure 1 The target object recognition method of the present application comprises the following steps:

[0073] In the existing target object detection, the image detection model generally only detects the target object in a single region of the vehicle end image. The size of the single region is various, and the input image detection model is a detection region after resizing (such as 640x320) of the single region of the vehicle end image. The resize() function is used to modify the picture size or adjust the picture scale to be the same as the target size. In the face of multiple target object detection tasks, the existing image detection model needs to increase the number of batch size. Since the batch size is not the larger the better (a better batch size can make the server GPU run at full capacity and make the model training and use speed optimal), under the limited vehicle end resources, constantly increasing the batch size will make the vehicle end server load larger and larger, which is easy to cause server response delay, even crash and unable to provide services, affect the normal driving of the driver, cause sudden accidents, etc. In order to obtain better detection effect, the corresponding detection region needs to be cropped according to the imaging region scale of the target object in the image, so that it is maintained within the better detection scale of the image detection model. However, under the limited detection region scale, the existing image cropping method is relatively random, and it is difficult to identify the more important target objects in the vehicle field of view.

[0074] Through the target object recognition method of the present application, under the limited vehicle end resources and the limited detection scale of the image detection model, the region priority can be determined according to the priority of the target object, the vehicle end image can be cropped and spliced based on the region priority, and the cropped and spliced image can be used as the input of the image detection model which can perform multi-region detection, so as to realize the multi-region simultaneous detection of as many and high-priority target objects as possible in the limited detection region, thereby maintaining the stability and diversity of the detection region, making different target objects have the possibility of being selected, and without consuming additional vehicle end resources.

[0075] Because the multi-region detection scales of the target objects corresponding to the input of the image detection model are different, the accuracy of the output of the corresponding image detection model is also different. For different region detection scales, the output accuracy is similar to a normal distribution, that is, it is not easy to detect if the target object is too large or too small. Therefore, the target object recognition method of the present invention aims to place the target objects in multiple regions within a better detection range. Thus, how to make the scales of as many higher-priority target objects as possible within the optimal detection scale range of the image detection model becomes the objective of the target object recognition method of the present invention.

[0076] Step S101: Obtain the current priority of multiple target objects to be detected included in the vehicle image; wherein the current priority is determined based on the vehicle's driving data.

[0077] In this embodiment of the invention, the target object mainly refers to some fixed targets within the vehicle's field of vision, such as traffic lights, road bollards, and landmark buildings with relatively fixed features.

[0078] In this embodiment of the invention, the vehicle's driving data includes map data, and the map data includes the target object O to be detected within the vehicle's field of vision. i Location and route information.

[0079] In this embodiment of the invention, based on the target object O to be detected i Location and route information are used to determine the target object O to be detected. i The current priority of Prior(O) i Alternatively, the target object O within the vehicle's field of vision can be determined based on the type of detection task and the application scenario. i The current priority of Prior(O) i (); where the vehicle's field of view corresponds to the vehicle-side image.

[0080] In this embodiment of the invention, the priority determination principles for different types of target objects are different, and the corresponding priority determination algorithms are different. For example, if the target object is a traffic light, the priority determination algorithm is related to the positional and logical relationship of the traffic light.

[0081] Step S102: Based on the target input scale of the pre-trained image detection model and the current priority, determine multiple preprocessing regions of the vehicle-end image; wherein, the preprocessing region includes one or more of the target objects to be detected.

[0082] In the embodiment of the present application, the pre-trained image detection model can realize multi-region detection of multiple target objects in the vehicle end image in the limited detection region without consuming additional vehicle end resources. The target input scale is a detection scale that the pre-trained image detection model has determined and corresponds to a detection effect of the image detection model. The target input scale is an interval value, so that the model input determined according to the multiple cropping regions corresponding to one or more target objects to be detected is within the interval value. The multiple cropping regions are determined by cropping the multiple preprocessing regions based on the target objects to be detected. The model input is the result of splicing the multiple cropping regions, and the model input corresponds to the limited detection region.

[0083] In the embodiment of the present application, the preprocessing region includes a first type of preprocessing region and a second type of preprocessing region. The first type of preprocessing region is a stable priority region PriorArea p , and the second type of preprocessing region is a dynamic switching region SwitchArea q .

[0084] In the embodiment of the present application, as shown in Figure 2 , the training method of the image detection model of the present application includes the following steps:

[0085] Step S201, modifying the input scale of the image detection model.

[0086] In the embodiment of the present application, the input scale of the image detection model is changed by using the region construction method, so that the limited detection region under the existing single-region detection can realize multi-region detection of multiple target objects, that is, single-region detection of a vehicle end image is changed to multi-region detection of a vehicle end image. The input of the changed image detection model corresponds to the target input scale.

[0087] The target input scale indicates the pixel size of the detection region of the changed image detection model. The detection region includes multiple cropping regions corresponding to multiple target objects to be detected. Accordingly, the preprocessing region of the vehicle end image is cropped according to the target input scale, so that the splicing result of the multiple cropping regions meets the input of the image detection model, and then multi-region detection of multiple target objects to be detected in the vehicle end image is performed. One cropping region can include one or more target objects to be detected.

[0088] The cropping size of the to-be-detected target object in the preprocessing area is Size crop. One vehicle-end image corresponds to multiple preprocessing areas, and the area sizes of the multiple preprocessing areas can be the same or different. The cropping sizes of the to-be-detected target objects are the same, that is, the to-be-detected target objects in the preprocessing areas are cropped according to the cropping size, and then the cropping results are spliced as the input of the image detection model. The input of the image detection model corresponds to the detection area of the image detection model, and the multi-area detection of the to-be-detected target object can be realized. For example, the target input size of the image detection model is Size input = N*N, N*N can be the pixel size (such as length* width) of the target input size. In order to make the cropping area and the target input size Size input satisfy the expected proportional relationship, the cropping size can be determined according to the number of cropping areas included in the detection area, specifically:

[0089] Suppose that the image detection model detects 2 areas at the same time, correspondingly, the detection area includes 2 cropping areas, and the cropping size Size crop can be Size crop = N / 2*N. The to-be-detected target objects in the preprocessing areas are cropped according to the cropping size N / 2*N, and 2 N / 2*N cropping areas are obtained. The 2 cropping areas are spliced as the detection area, so that the multi-area detection (that is, multi-area detection) can be realized at one time. By analogy, suppose that the image detection model detects 4 areas at the same time, and each cropping size can be Size crop = N / 2*N / 2.

[0090] In the embodiment of the application, the target input size can also be represented by the format of the image data channel, which can be selected according to actual needs.

[0091] In step S202, the image detection model is trained.

[0092] In the embodiment of the application, the image detection model is trained according to the input size of the modified image detection model, so that the trained image detection model can perform multi-area detection.

[0093] In the embodiment of the application, by using the training method of the image detection model, the image detection model can be improved to adapt to the multi-area detection of the target object of the vehicle-end image, and no additional vehicle-end resources are consumed, thereby improving the detection efficiency of the model.

[0094] In the embodiment of the application, as shown in Figure 3 The determination method of the preprocessing area of the application includes the following steps:

[0095] In step S301, the target area size of the to-be-detected target object is determined according to the target input size.

[0096] In step S3011, the real size of the target object to be detected is determined according to the driving data of the vehicle.

[0097] In the embodiment of the present application, the vehicle-side image corresponds to a camera coordinate system of the vehicle, the map data of the driving data of the vehicle includes the real size of the target object to be detected, and the position information includes a position coordinate.

[0098] In step S3012, the target region scale of the target object to be detected is determined according to the real size and the target input scale in the camera coordinate system.

[0099] In the embodiment of the present application, the target input scale indicates a target scale (i.e., an expected pixel size of the target object to be detected after being cropped before the image detection model is input) in a detection region of the image detection model that is most suitable for detecting the target object to be detected, that is, a cropping scale Size crop of the target object to be detected, the target input scale corresponds to a scale of the detection region of the image detection model, the detection region corresponds to a plurality of target scales, the plurality of target scales are the same, and one target scale can include one or more target objects to be detected. The target scale is determined according to the number of target scales corresponding to the detection region of the pre-trained image detection model. For example, the target input scale is N*N, the number of target scales corresponding to the detection region of the pre-trained image detection model is 4, and the target scale is determined as N / 2*N / 2.

[0100] In the camera coordinate system, the target region scale of the target object to be detected in the vehicle-side image is determined according to the real size, the position coordinate, and the target scale of the target object to be detected based on the camera imaging principle; wherein the target scale of the target object to be detected is determined after the image detection model is trained.

[0101] Further, the target region scale of the target object to be detected is determined according to the real size, the camera projection distance, and the target scale of the target object to be detected. i .

[0102] In step S302, a candidate region set of a plurality of target objects to be detected is determined according to a preset edge scale and the target region scale; wherein the candidate region set includes a plurality of candidate region frames, and one or more target objects to be detected are included in the candidate region frames.

[0103] In the embodiment of the present application, the region cropping scale of the target object to be detected is determined by using the padding filling method according to the preset edge scale and the target region scale, so as to ensure the information integrity of the feature map. On the basis of the region cropping scale, the corresponding candidate region frame is generated, so as to obtain as many target objects to be detected and their surrounding environment as possible, so as to generate the candidate region and form the candidate region set including a plurality of target objects to be detected.

[0104] In the embodiments of the present application, as shown in Figure 4 the method for determining the candidate region of the present application comprises the following steps:

[0105] In step S401, a plurality of candidate region frames corresponding to the target object to be detected are generated according to the preset edge scale and the target region scale.

[0106] In the embodiments of the present application, in order to avoid the target object to be detected from being truncated due to falling into the region edge, the edge scale is set to determine the region clipping scale Scale i , as shown in the following formula:

[0107] Scale i = S i -M i

[0108] wherein, the preset edge scale M i is 10px, or the preset edge scale M i may be selectively set according to the requirement.

[0109] Further, different candidate region frames B1, B2, …, B i are generated according to the region clipping scale Scale n of the target object to be detected; wherein the target object to be detected is included in the candidate region frame.

[0110] In step S402, the plurality of candidate region frames are sorted according to the scale of the candidate region frames.

[0111] In the embodiments of the present application, the plurality of candidate region frames are sorted according to the order from small to large scale, and the candidate region frame with the smallest scale is determined as the minimum candidate region frame.

[0112] In step S403, according to the sorting result, the plurality of candidate regions of the target object to be detected included in the vehicle end image are determined to form the candidate region set.

[0113] In the embodiments of the present application, steps S4031-S4034 are executed cyclically from the minimum candidate region frame until the candidate region includes the maximum number of target objects to be detected.

[0114] In step S4031, the target object to be detected corresponding to the candidate region frame is determined to be the nearest one or more other target objects to be detected according to the scale of the candidate region frame.

[0115] In the embodiment of the present application, the minimum region frame is taken as the current candidate region frame, the size of the current candidate region frame is determined, and the current target object to be detected corresponding to the current candidate region frame is determined. According to the size of the current candidate region frame, the other target object to be detected closest to the current target object to be detected is determined.

[0116] In step S4032, it is determined whether the size of the region including the target object to be detected corresponding to the candidate region frame and the other target object to be detected is less than or equal to the size of the candidate region frame. If yes, the process proceeds to step S4033; if no, the process proceeds to step S4034.

[0117] In the embodiment of the present application, in the case where the other target object to be detected is one, it is determined whether the size of the region including the current target object to be detected, the other target object to be detected and the surrounding environment is less than or equal to the size of the current candidate region frame.

[0118] In the embodiment of the present application, in the case where the other target object to be detected is more than one, the plurality of other target objects to be detected are sorted, the other target object to be detected with the largest size is determined, and the size of the region including the current target object to be detected, the other target object to be detected with the largest size and the surrounding environment is compared with the size of the current candidate region frame. If the size of the region including the current target object to be detected, the other target object to be detected with the largest size and the surrounding environment is less than or equal to the size of the current candidate region frame, the other target object to be detected with the second largest size is determined according to the size sorting of the plurality of other target objects to be detected, and the size of the region including the current target object to be detected, the other target object to be detected with the largest size, the other target object to be detected with the second largest size and the surrounding environment is compared with the size of the current candidate region frame. The process is repeated in this way.

[0119] In step S4033, the other target object to be detected is added as the target object to be detected corresponding to the candidate region frame, and the process returns to step S4031.

[0120] In the embodiment of the present application, in the case where the size of the region including the target object to be detected corresponding to the candidate region frame and the other target object to be detected is less than or equal to the size of the candidate region frame, the other target object to be detected is added as the target object to be detected corresponding to the candidate region frame, and the loop comparison is continued to obtain the candidate region containing more target objects to be detected and / or the surrounding environment in turn.

[0121] In step S4034, the candidate region frame with any angle is taken as the candidate region.

[0122] In the embodiment of the present application, in the case that the region scale including the to-be-detected target object corresponding to the candidate region frame and other to-be-detected target objects is greater than the scale of the candidate region frame, the region corresponding to the candidate region frame of any angle is taken as a candidate region, and the candidate region includes the to-be-detected target object corresponding to the candidate region frame.

[0123] In the embodiment of the present application, steps S4031-S4034 are as follows:

[0124]

[0125] Wherein:

[0126] B(O) represents a set of to-be-detected target objects in the candidate region frame;

[0127] Num() represents the number of elements in the set B(O);

[0128] Box() represents a set of candidate regions;

[0129] Box(O)&Num(B(O))>Num(Box(O)) represents that the region including more to-be-detected target objects is selected.

[0130] In the embodiment of the present application, by using the candidate region determination method of the present application, different candidate region frames can be generated based on the region cropping scale, so that the candidate region frames can tentatively frame more to-be-detected target objects to obtain candidate regions, which facilitates the subsequent identification of the to-be-detected target objects and improves the identification amount and efficiency of the to-be-detected target objects.

[0131] Step S303, determining the plurality of pre-processing regions according to the current priority and target number of the to-be-detected target objects in the candidate region set.

[0132] In the embodiment of the present application, based on the difference between detection stability and dynamics, the candidate regions in the candidate region set Box m are distinguished. The priority of the stable candidate region is usually determined and relatively fixed, while the priority of the dynamic candidate region can be adjusted multiple times according to the detection result. m For each candidate region included in the candidate region set Box

[0133] Step S3031, determining the candidate priority of each candidate region included in the candidate region set according to the current priority and target number of the to-be-detected target objects included in the candidate region.

[0134] In the embodiments of the present application, the candidate priority Prior(Box) of each candidate region is generally determined by the priority of one or more target objects to be detected in the candidate region. When the candidate priority Prior(Box) of the candidate region is determined for the first time, the priority of the target object to be detected is the initial current priority. When the candidate priority Prior(Box) of the candidate region is determined based on the detection result, the priority of the target object to be detected is the adjusted current priority.

[0135] In the embodiments of the present application, when the target number of the target object to be detected included in the candidate region is one, the candidate priority is the current priority of the target object to be detected.

[0136] When the target number of the target object to be detected included in the candidate region is multiple,

[0137] The candidate priority is the sum of the current priorities of the multiple target objects to be detected. The following formula is shown:

[0138]

[0139] The above formula indicates that:

[0140] The candidate priority of the candidate region is determined according to the sum of the priorities of one or more target objects to be detected in the candidate region.

[0141] In step S3032, the candidate region with the highest candidate priority is determined as the first type of preprocessing region, and the remaining candidate regions are determined as the second type of preprocessing region.

[0142] In the embodiments of the present application, the first type of preprocessing region (i.e., the stable priority region PriorArea p ) is determined by the following formula:

[0143] PriorArea p ={Area∈{Box∈Box m |Max(Num(Box(O)))}|Max(Prior(Area))}

[0144] The above formula indicates that:

[0145] The candidate region with the largest target number of target objects to be detected is selected, and the region priority is determined. The corresponding candidate region is the stable priority region PriorArea p .

[0146] The second type of preprocessing region (i.e., the dynamic switching region SwitchArea q ) is determined by the following formula:

[0147]

[0148] The above formula represents: in addition to the stable priority area PriorArea p , in the case that the maximum current priority of the target objects included in the plurality of candidate areas is the same, the candidate area with the most target objects is selected as the dynamic switching area SwitchArea q .

[0149] In an embodiment of the present application, or in the case that the number of target objects included in the plurality of candidate areas is the same, the candidate area with the maximum current priority of the target objects to be detected in the candidate area is selected as the dynamic switching area SwitchArea q ; or, all the remaining candidate areas are taken as the dynamic switching area SwitchArea q .

[0150] In an embodiment of the present application, more target objects to be detected are tentatively framed by the candidate area frame, so that the target objects to be detected in the pre-processing area are more evenly distributed.

[0151] In an embodiment of the present application, by the determination method of the pre-processing area of the present application, the target scale of the target object to be detected can be determined based on the target input scale of the pre-trained image detection model, so as to determine the candidate area frame to frame as many target objects to be detected as possible, generate a candidate area set, and determine the stable priority area PriorArea p and the dynamic switching area SwitchArea q according to the priority and number of the target objects to be detected, so as to realize the detection of as many target objects to be detected as possible with high priority by the image detection model, without consuming additional vehicle-side resources, and ensuring the stability of the vehicle-side service and user experience.

[0152] Step S103: processing the vehicle-side image according to the plurality of pre-processing areas, and taking the processed vehicle-side image as the input of the image detection model to identify the target object.

[0153] In an embodiment of the present application, the processing of the vehicle-side image can be cropping, splicing, etc. The vehicle-side image is cropped to obtain a plurality of cropped areas corresponding to the target objects to be detected included in the pre-processing area, the plurality of cropped areas are spliced to generate the input of the image detection model, and the target object is identified according to the output of the image detection model.

[0154] In an embodiment of the present application, as Figure 5 shown, the determination method of the input of the image detection model of the present application includes the following steps:

[0155] Step S501, according to the plurality of preprocessing areas and the preset cropping scale, cropping the vehicle end image to obtain a plurality of cropped areas.

[0156] In the embodiment of the application, the cropping scale is determined by the pre-trained image detection model, and the preprocessing area is cropped according to the cropping scale to obtain a plurality of cropped areas.

[0157] Step S502, splicing the plurality of cropped areas.

[0158] In the embodiment of the application, the splicing method can be row splicing, that is, without changing the width, splitting the height, and splicing the height split area into a detection area by row. As shown in the following formula:

[0159] DetectArea=vconcat(PriorArea p ,SwitchArea q ).

[0160] The above formula indicates that the cropped areas obtained by cropping the preprocessing area are spliced, that is, the cropped areas obtained by respectively cropping the stable priority area and the dynamic switching area are spliced.

[0161] Step S503, determining the input of the image detection model according to the splicing result.

[0162] In the embodiment of the application, the splicing result corresponds to the detection area of the image detection model.

[0163] Further, the image processing model independent of the image detection model can be used to determine the splicing result of the vehicle end image. The image processing model performs the preprocessing, cropping and splicing of the vehicle end image according to the obtained vehicle end image, and outputs the splicing result as the input of the image detection model.

[0164] If the single area is detected respectively according to the prior art to identify a plurality of target objects, the batch size needs to be increased, which causes great consumption of vehicle end resources and greatly increases the operation time. In the embodiment of the application, as shown in the formula (1), without increasing the consumption of vehicle end resources, the multi-area detection of a plurality of target objects can be realized corresponding to the existing limited detection area. Figure 6

[0165] ​In the embodiment of the present application, the method for determining the input of the image detection model can crop multiple target objects in multiple preprocessing regions, and splice the cropped regions to input the image detection model, so that multi-region detection of multiple target objects in a limited detection region can be realized, the detection efficiency of the model is greatly improved, and the vehicle-side resources are not additionally consumed, so that the stability of the vehicle-side service is ensured.

[0166] In the embodiment of the present application, as shown in Figure 7 The method for adjusting the priority of the target object comprises the following steps:

[0167] In step S701, the priority of the target object corresponding to the second type of detection region is adjusted according to the current detection result of the image detection model.

[0168] In the embodiment of the present application, after the spliced result is input into the image detection model, the first detection result of the image detection model can be obtained. However, the detection result of the stable priority region is usually accurate, while the detection result of the dynamic switching region is usually less accurate, so the priority of the target object in the dynamic switching region needs to be adjusted. The evaluation criteria of the detection result are determined in advance, and can be determined according to the detection task, the detection scene, etc. For example, the priority of the target object is adjusted due to low detection score, unstable detection, etc. The adjustment criteria can be to reduce the priority of the target object with unstable detection, so as to improve the detection effect of the dynamic switching region.

[0169] In step S702, the target detection result of the image detection model is determined again according to the target input scale of the image detection model and the current priority.

[0170] In the embodiment of the present application, after the priority of the target object is adjusted, the method for identifying the target object of the present application is executed again according to the adjusted priority of the target object, and the iteration is repeated until the best detection result is obtained.

[0171] In the embodiment of the present application, the method for adjusting the priority of the target object of the present application can adjust the priority of the target object to re-detect when a better detection result is not obtained, and the detection result is optimized by continuously iterating.

[0172] In the embodiment of the present application, as shown in Figure 8 The method for determining the output of the image detection model comprises the following steps:

[0173] Step S801, normalizing the target detection result of the image detection model.

[0174] In the embodiment of the present application, since the scale of the target object corresponding to the cropped region may not be the same as the original vehicle end image, in order to facilitate the viewing of the output, the target detection result is normalized before output to restore the scale of the original vehicle end image and then output.

[0175] After the detection of the region is completed, since the scales of different regions may not be consistent, the normalized detection frame is uniformly subjected to NMS to realize multi-region fusion after the normalized detection frame is normalized to the scale of the original image according to each cropped region, and then the final multi-region detection result is output.

[0176] Step S802, determining the output of the image detection model according to the processing result of the normalization processing.

[0177] In the embodiment of the present application, since the frame selection content of the plurality of candidate region frames may overlap, in order to ensure the accuracy of the final output, the multi-region fusion of the target detection result subjected to the normalization processing is realized by using the NMS algorithm, which is equivalent to the de-overlapping processing, and ensures that only one object is reserved in the image output by the model.

[0178] Step S803, identifying the target object according to the output of the image detection model.

[0179] In the embodiment of the present application, the image detection model outputs the multi-region detection result, that is, each target object in the vehicle end image.

[0180] In the embodiment of the present application, by using the determination method of the output of the image detection model, the target detection result of the image detection model can be normalized to correspond to the size of the original vehicle end image, which facilitates the viewing of the output, and by using the NMS fusion processing, only one object is reserved in the image, the interference of the overlapping region on the output is excluded, the accuracy of the output is ensured, and the recognition accuracy of the image detection model is improved.

[0181] In the embodiment of the present application, the current priority of the plurality of target objects to be detected included in the vehicle terminal image is acquired, wherein the current priority is determined according to the driving data of the vehicle; a plurality of detection regions of the vehicle terminal image are determined according to the target input scale of the pre-trained image detection model and the current priority, wherein the detection region includes one or more target objects to be detected; the vehicle terminal image is processed according to the plurality of detection regions, and the processed vehicle terminal image is taken as the input of the image detection model to identify the target object, etc. Steps can be implemented in a limited detection region to cover more number and higher priority target objects, and without consuming additional vehicle terminal resources, thereby realizing multi-region detection of target objects, improving user experience, and ensuring driving safety and stability of vehicle terminal services.

[0182] Figure 9 is a schematic diagram of the main modules of the target object identification device according to the embodiment of the present application, as shown in Figure 9 The target object identification device 900 of the present application includes:

[0183] The acquisition module 901 is configured to acquire the current priority of the plurality of target objects to be detected included in the vehicle terminal image, wherein the current priority is determined according to the driving data of the vehicle.

[0184] In the embodiment of the present application, the acquisition module 901 is configured to acquire the current priority (i.e. initial priority) of the plurality of target objects to be detected included in the vehicle terminal image. The driving data of the vehicle includes map data, and the map data includes the position information and route information of the target object O i to be detected within the field of view of the vehicle. According to the position information and route information of the target object O i to be detected, the current priority Prior(O i ) of the target object O i to be detected is determined, or the current priority Prior(O i ) of the target object O i to be detected within the field of view of the vehicle can be determined according to the type and application scenario of the detection task.

[0185] The preprocessing module 902 is configured to determine a plurality of preprocessing regions of the vehicle terminal image according to the target input scale of the pre-trained image detection model and the current priority, wherein the preprocessing region includes one or more target objects to be detected.

[0186] In the embodiment of the present application, the pre-trained image detection model can realize multi-region detection of multiple target objects in the vehicle end image in the limited detection area without consuming additional vehicle end resources. The preprocessing module 902 is configured to determine multiple preprocessing regions of the vehicle end image according to a target input scale of the pre-trained image detection model and a current priority of the target object to be detected. The target input scale is a detection scale that the pre-trained image detection model has determined to correspond to a better detection effect of the image detection model. The target input scale is an interval value, so that the model input determined according to the multiple cropping regions corresponding to one or more target objects to be detected is within the interval value. The multiple cropping regions are determined by cropping the multiple preprocessing regions based on the target object to be detected. The model input is the result of splicing the multiple cropping regions, and the model input corresponds to a limited detection area.

[0187] In the embodiment of the present application, the preprocessing region includes a first type of preprocessing region and a second type of preprocessing region. The first type of preprocessing region is a stable priority region PriorArea p , and the second type of preprocessing region is a dynamic switching region SwitchArea q .

[0188] The recognition module 903 is configured to process the vehicle end image according to the multiple preprocessing regions, and take the processing result as the input of the image detection model to recognize the target object

[0189] In the embodiment of the present application, the processing of the vehicle end image can be cropping, splicing, etc. The recognition module 903 crops the vehicle end image to obtain multiple cropping regions corresponding to the target object to be detected included in the preprocessing region, splices the multiple cropping regions to generate the input of the image detection model, and recognizes the target object according to the output of the image detection model.

[0190] In the embodiment of the present application, through the modules such as the acquisition module, the preprocessing module and the recognition module, more number and higher priority target objects can be covered in the limited detection area without consuming additional vehicle end resources, so as to realize multi-region detection of the target object, improve user experience, and ensure driving safety and stability of the vehicle end service.

[0191] Figure 10 An exemplary system architecture diagram of a target object recognition method or a target object recognition device suitable for application in the embodiment of the present application is shown, as shown in Figure 10 The exemplary system architecture of the target object recognition method or the target object recognition device in the embodiment of the present application includes:

[0192] As Figure 10As shown, the system architecture 1000 can include detection devices 1001, 1002, 1003, a network 1004, and a server 1005. The network 1004 is a medium for providing a communication link between the detection devices 1001, 1002, 1003 and the server 1005. The network 1004 can include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0193] The detection devices 1001, 1002, 1003 interact with the server 1005 through the network 1004 to collect or send messages, etc.

[0194] The detection devices 1001, 1002, 1003 can be various electronic devices with detection functions, including but not limited to laser radar sensors, cameras, etc.

[0195] The server 1005 can be a server providing various services, such as a background management server providing support for the collected vehicle-side images sent by the detection devices 1001, 1002, 1003. The background management server can analyze and process the collected data including the vehicle-side images of multiple target objects to be detected, etc., and output the processing results (such as the recognition results of the target objects).

[0196] It should be noted that the target object recognition method provided by the embodiments of the present application is generally executed by the server 1005, and correspondingly, the target object recognition apparatus is generally arranged in the server 1005.

[0197] It should be understood that, Figure 10 The number of detection devices, networks, and servers in the system architecture 1000 is only illustrative. According to the needs of implementation, there can be any number of detection devices, networks, and servers.

[0198] Figure 11 is a structural schematic diagram of a computer system of a terminal device or a server suitable for implementing the embodiments of the present application, as Figure 11 As shown, the computer system 1100 of the terminal device or the server of the embodiments of the present application includes:

[0199] A central processing unit (CPU) 1101 can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 1102 or programs loaded from a storage portion 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the system 1100 are also stored. The CPU 1101, the ROM 1102, and the RAM 1103 are connected to each other through a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0200] The following components are connected to the I / O interface 1105: an input part 1106 including a keyboard, a mouse, etc.; an output part 1107 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 1108 including a hard disk, etc.; and a communication part 1109 including a network interface card such as a LAN card, a modem, etc. The communication part 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I / O interface 1105 as necessary. A removable media 1111 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 1110 as necessary, so that a computer program read out therefrom is installed in the storage part 1108 as necessary.

[0201] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication part 1109, and / or installed from the removable media 1111. When the computer program is executed by the central processing unit (CPU) 1101, the above-described functions defined in the system of the present disclosure are executed.

[0202] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0203] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0204] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. The described modules can also be arranged in a processor, for example, a processor can be described as including an acquisition module, a preprocessing module, and an identification module. In some cases, the names of these modules do not constitute a limitation on the modules themselves, for example, the preprocessing module can also be described as "a module that determines a plurality of preprocessing regions of the vehicle-side image according to a target input scale of a pre-trained image detection model and the current priority".

[0205] As another aspect, the present application also provides a computer readable medium, which can be included in the device described in the above embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, which, when executed by the device, cause the device to include: acquiring a current priority of a plurality of target objects included in a vehicle-side image; wherein the current priority is determined according to driving data of the vehicle; determining a plurality of preprocessing regions of the vehicle-side image according to a target input scale of a pre-trained image detection model and the current priority; wherein the preprocessing region includes one or more of the target objects to be detected; and processing the vehicle-side image according to the plurality of preprocessing regions to determine a plurality of detection regions as inputs of the image detection model to identify the target objects.

[0206] According to the technical solutions of the embodiments of the present application, through a plurality of algorithms (including a target object candidate region generation algorithm, a multi-region priority determination algorithm, a detection region construction algorithm, and a multi-region detection result fusion algorithm), more target objects with higher priority can be included in a limited detection region without consuming additional vehicle-side resources, thereby improving the accuracy of the identification result.

[0207] According to the technical solutions of the embodiments of the present application, the existing image detection model is improved so that the image detection model can perform multi-region detection, and then the detection scale of the target object is determined according to the performance of the trained image detection model, and then the region with higher priority is selected in the limited detection region based on the limited batch size, and finally the multi-region detection result is fused to output the identification result.

[0208] According to the technical scheme of the embodiment of the present application, the existing image detection model is improved, so that the image detection model can perform multi-region detection, and then the target scale of the target object to be detected is determined according to the target input scale of the trained image detection model; the target region scale of the target object to be detected is determined according to the target scale of the target object to be detected; on the basis of the target region scale, a candidate region frame is constructed, so as to determine a candidate region set starting from the smallest candidate region frame, for the purpose of including as many target objects to be detected as possible; then the priority of the candidate region is determined according to the initial priority of the target object to be detected and the target quantity, to obtain various types of pre-processing regions (including stable priority regions and dynamic switching regions); the pre-processing region is cropped and spliced to generate the input of the image detection model; the priority of the target object is adjusted according to the detection result, to obtain the final target detection result; the target detection result is normalized and fused by NMS, so as to obtain the output of the image detection model, to determine the multi-region detection result of the vehicle-side image including multiple target objects to be detected, so as to realize the multi-region detection of the vehicle-side image including multiple target objects to be detected without increasing the batch size and without occupying additional vehicle-side resources, to ensure the region stability of the multi-region detection and to meet the target diversity of the multi-region detection.

[0209] According to the technical scheme of the embodiment of the present application, more target objects of a higher priority can be covered in a limited detection region without consuming additional vehicle-side resources, so as to realize the multi-region detection of the target objects, to improve the user experience and to ensure the driving safety and the stability of the vehicle-side service.

[0210] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made depending on design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying a target object, characterized in that, include: The current priority of multiple target objects to be detected in the vehicle image is obtained; wherein the current priority is determined based on the vehicle's driving data. Based on the target input scale of the pre-trained image detection model and the current priority, multiple preprocessing regions of the vehicle-end image are determined; wherein, the preprocessing region includes one or more of the target objects to be detected, and the target input scale is an interval value; The vehicle-end image is processed according to multiple preprocessing regions, and the processing result is used as the input of the image detection model to identify the target object; The step of determining multiple preprocessing regions of the vehicle-end image based on the target input scale of the pre-trained image detection model and the current priority includes: Based on the target input scale, determine the target region scale of the target object to be detected; Based on a preset edge scale and the target region scale, a set of candidate regions for multiple target objects to be detected is determined; wherein, the set of candidate regions includes multiple candidate region boxes, and each candidate region box includes one or more target objects to be detected; The multiple preprocessing regions are determined based on the current priority and number of the target objects to be detected within the candidate region set.

2. The method according to claim 1, characterized in that, The step of determining a set of candidate regions for multiple target objects to be detected based on a preset edge scale and the target region scale includes: Based on the preset edge scale and the target region scale, multiple candidate region boxes corresponding to the target object to be detected are generated. The candidate region boxes are sorted according to their scale. Based on the sorting results, multiple candidate regions of all the target objects to be detected included in the vehicle-end image are determined and formed into the candidate region set.

3. The method according to claim 2, characterized in that, The step of determining multiple candidate regions of all the target objects to be detected included in the vehicle-end image includes: Starting from the smallest candidate region bounding box, repeat the following steps until the candidate region contains the maximum number of target objects to be detected: A1: Based on the scale of the candidate region box, determine one or more other target objects that are closest to the target object to be detected corresponding to the candidate region box; A2: Determine whether the scale of the region including the target object to be detected corresponding to the candidate region box and other target objects to be detected is less than or equal to the scale of the candidate region box. If yes, proceed to step A3; if no, proceed to step A4. A3: Add the other target objects to be detected as the target objects to be detected corresponding to the candidate region box, and go to step A1; A4: Select the candidate region box at any angle as the candidate region.

4. The method according to claim 1, characterized in that, The step of determining the multiple preprocessing regions based on the current priority and number of the target objects to be detected within the candidate region set includes: Based on the current priority and number of the target objects to be detected included in the candidate regions, the candidate priority of each candidate region included in the candidate region set is determined; The candidate region with the highest priority is identified as the first type of preprocessing region, and the remaining candidate regions are identified as the second type of preprocessing regions.

5. The method according to claim 4, characterized in that, The step of determining the candidate priority of each candidate region in the candidate region set based on the current priority and number of the target objects to be detected included in the candidate region includes: When the number of targets to be detected included in the candidate region is one, the candidate priority is the current priority of the target object to be detected; When the number of targets to be detected included in the candidate region is multiple, the candidate priority is the sum of the current priorities of the multiple targets to be detected.

6. The method according to claim 1, characterized in that, The vehicle-side image corresponds to the camera coordinate system of the vehicle; determining the target region scale of the target object to be detected based on the target input scale includes: Based on the vehicle's driving data, determine the physical dimensions of the target object to be detected; In the camera coordinate system, the target area scale of the target object to be detected is determined based on the actual object size and the target input scale.

7. The method according to claim 1, characterized in that, The step of processing the vehicle-end image based on multiple preprocessed regions and using the processing result as input to the image detection model includes: The vehicle-end image is cropped according to the multiple preprocessing regions and the preset cropping scale to obtain multiple cropping regions; The multiple cut areas are then joined together; Based on the stitching results, the input to the image detection model is determined.

8. The method according to claim 4, characterized in that, Also includes: Based on the current detection results of the image detection model, the priority of the target object corresponding to the second type of preprocessing region is adjusted; Based on the adjusted priority of the target object, determine the current priority of the target object; Based on the adjusted current priority of the target object, the process of determining the target detection result of the image detection model by re-executing the process based on the target input scale of the image detection model and the current priority is performed.

9. The method according to claim 8, characterized in that, Also includes: The target detection results of the image detection model are normalized. The output of the image detection model is determined based on the normalization processing result; The target object is identified based on the output of the image detection model.

10. The method according to any one of claims 1-9, characterized in that, The image detection model is trained based on a modified input scale; wherein the modified input scale corresponds to multi-region detection of multiple target objects.

11. A target object identification device, characterized in that, include: An acquisition module is used to acquire the current priority of multiple inherent targets to be detected included in the vehicle-side image; wherein the current priority is determined based on the vehicle's driving data; The preprocessing module is used to determine multiple preprocessing regions of the vehicle-end image based on the target input scale of the pre-trained image detection model and the current priority; wherein, the preprocessing region includes one or more of the inherent targets to be detected, and the target input scale is an interval value; The recognition module is used to process the vehicle-end image based on multiple preprocessed regions, and use the processing result as input to the image detection model to recognize the inherent target; The preprocessing module is also used for: Based on the target input scale, determine the target region scale of the target object to be detected; Based on a preset edge scale and the target region scale, a set of candidate regions for multiple target objects to be detected is determined; wherein, the set of candidate regions includes multiple candidate region boxes, and each candidate region box includes one or more target objects to be detected; The multiple preprocessing regions are determined based on the current priority and number of the target objects to be detected within the candidate region set.

12. An electronic device for identifying a target object, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-10.

13. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-10.

Images