Target object detection model determination method after training

Abstract

The embodiment of the invention provides a method for determining a trained target object detection model, and relates to the technical field of road target detection. The method comprises the following steps: acquiring a road image data set of a vehicle under different road conditions, and performing rotation transformation on each road image in the road image data set to obtain an input image; inputting the input image into the initial target object detection model, and performing illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model to obtain a first target image; identifying the first target image based on a target object detection layer in the initial target object detection model, and determining actual label information of a target object in the first target image; and according to the real label information of the target object and the actual label information of the target object, training an initial target object detection model to obtain a trained target object detection model. With the adoption of the technical scheme, the target object can be accurately detected in a low-illumination scene, so that the influence of light and shade is avoided.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of road target detection, and in particular to a method for determining a trained target object detection model. Background Art

[0002] At present, autonomous driving technology is generally at the L2 stage, mainly including ACC, ICC, AEB and other functions. These technologies are usually divided into three levels: perception layer, decision layer and execution layer. Among them, the perception layer is the input end of the intelligent driving function, and the target detection algorithm is an important component of it.

[0003] However, object detection algorithms are easily affected by environmental factors, especially changes in light brightness.

[0004] Therefore, there is an urgent need for a method to determine a trained target object detection model that can accurately detect target objects in low-light scenes and avoid the influence of light brightness. Summary of the invention

[0005] In order to solve the above technical problem or at least partially solve the above technical problem, an embodiment of the present disclosure provides a method for determining a trained target object detection model.

[0006] A first aspect of an embodiment of the present disclosure provides a method for determining a trained target object detection model, the method comprising:

[0007] Acquire a road image dataset of a vehicle under different road conditions, and perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to characterize the position information and category information of each target object;

[0008] Repeat the following steps until the preset conditions are met;

[0009] Inputting the input image into an initial target object detection model, and performing illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model to obtain a first target image;

[0010] Recognize the first target image based on the target object detection layer in the initial target object detection model, and determine actual label information of the target object in the first target image;

[0011] The initial target object detection model is trained according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0012] In one example, performing illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model to obtain the first target image includes:

[0013] Based on the self-calibration module in the illumination enhancement layer, self-calibrate the current illumination intensity of the input image to obtain a calibrated illumination intensity;

[0014] Based on the illumination estimation module in the illumination enhancement layer, the calibrated illumination intensity is iterated to obtain a first target image.

[0015] In one example, the iterating the calibrated illumination intensity based on the illumination estimation module in the illumination enhancement layer to obtain the first target image includes:

[0016] Based on the mapping function in the light estimation module, learning the residual between the calibrated light intensity and the target light intensity;

[0017] Based on the illumination estimation module, the first target image is obtained according to the residual and the calibrated illumination intensity.

[0018] In one example, performing a rotation transformation on each road image in the road image dataset to obtain an input image includes:

[0019] Acquire pixel points in each road image, and acquire first coordinate information of the pixel points;

[0020] According to the affine matrix, the first coordinate information of the pixel point is rotated and transformed to obtain the second coordinate information of the pixel point;

[0021] The input image is determined according to the second coordinate information of the pixel point.

[0022] In one example, before identifying the first target image based on the target object detection layer in the initial target object detection model and determining actual label information of the target object in the first target image, the method further includes:

[0023] Based on the target object detection layer in the initial target object detection model, the first target image is processed to obtain a perception feature image in a vertical direction and a perception feature image in a horizontal direction;

[0024] A processed first target image is determined according to the vertical perception feature image, the horizontal perception feature image and the first target image.

[0025] A second aspect of the embodiments of the present disclosure provides a target object detection method, which is applied to a trained target object detection model, and the method includes:

[0026] Obtaining a current road image of the vehicle, and inputting the road image into a trained target object detection model;

[0027] Based on the trained target object detection model, a target object in the current road image is determined.

[0028] A third aspect of an embodiment of the present disclosure provides a device for determining a trained target object detection model, the device comprising:

[0029] A first acquisition module is used to acquire a road image dataset of a vehicle under different road conditions, and to perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to represent the position information and category information of each target object;

[0030] Repeat the following steps until the preset conditions are met;

[0031] A processing module, configured to input the input image into an initial target object detection model, and perform illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model to obtain a first target image;

[0032] A recognition module, configured to recognize the first target image based on the target object detection layer in the initial target object detection model, and determine actual label information of the target object in the first target image;

[0033] The training module is used to train the initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0034] A fourth aspect of the embodiments of the present disclosure provides a target object detection device, which is applied to a trained target object detection model, and the device includes:

[0035] A second acquisition module is used to acquire a current road image of the vehicle and input the road image into a trained target object detection model;

[0036] The third determination module is used to determine the target object in the current road image based on the trained target object detection model.

[0037] A fifth aspect of an embodiment of the present disclosure provides an electronic device, comprising: a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the method of the first aspect above.

[0038] A sixth aspect of an embodiment of the present disclosure provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method of the first aspect described above can be implemented.

[0039] The disclosed embodiment provides a method for determining a trained target object detection model, the method comprising: obtaining a road image dataset of a vehicle under different road conditions, performing a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; the real label information is used to characterize the position information and category information of each target object; repeatedly performing the following steps until a preset condition is met; inputting the input image into an initial target object detection model, performing illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model to obtain a first target image; recognizing the first target image based on the target object detection layer in the initial target object detection model to determine the actual label information of the target object in the first target image; training the initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model. By adopting this technical solution, the target object can be accurately detected in a low-light scene, thereby avoiding the influence of light brightness. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Figure 1 is a flowchart of a method for determining a trained target object detection model provided by an embodiment of the present disclosure;

[0043] Figure 2a is a flowchart of a method for determining a trained target object detection model provided by an embodiment of the present disclosure;

[0044] Figure 2b is a structural diagram of a CA attention mechanism provided by an embodiment of the present disclosure;

[0045] Figure 2c is a schematic diagram of the structure of a trained target object detection model provided by an embodiment of the present disclosure;

[0046] Figure 3 is a flow chart of a target object detection method provided by an embodiment of the present disclosure;

[0047] Figure 4 is a structural schematic diagram of a device for determining a trained target object detection model provided by an embodiment of the present disclosure;

[0048] Figure 5 is a structural schematic diagram of a target object detection device provided by an embodiment of the present disclosure;

[0049] Figure 6 It is a structural schematic diagram of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0050] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0052] Figure 1 1 is a flow chart of a method for determining a trained target object detection model provided by an embodiment of the present disclosure, and the method can be performed by an electronic device. The electronic device can be exemplarily understood as a device such as a mobile phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, etc. Figure 1 As shown, the method provided in this embodiment includes the following steps:

[0053] S101. Obtain a road image dataset of a vehicle under different road conditions, and perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to characterize the position information and category information of each target object.

[0054] In one example, different road conditions represent conditions on different roads, or conditions on the same road at different time periods. Furthermore, the illumination intensities corresponding to different time periods are different, and the vehicle conditions on the road at different time periods are also different.

[0055] In one example, a rotation transformation is performed on each road image in a road image dataset, which can change the position information of each target object in each road image, thereby increasing the data volume of the road image dataset and improving the detection capability of the initial target object detection model.

[0056] In this embodiment, each target object in each road image has real label information, which can be manually marked by the user in advance. The real label information includes the location information and category information of each target object. Specifically, the location information can be the coordinate information of the target object in the road image, and the category information can be a truck, a passenger car, or a pedestrian. For example, the category information of a truck is recorded as A, the category information of a passenger car is recorded as B, and the category information of a pedestrian is recorded as C.

[0057] S102, repeat the following steps until the preset conditions are met.

[0058] In one example, the preset condition may be the number of executions, for example, the number of executions may be 100 times.

[0059] S103: Input the input image to an initial target object detection model, and perform illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model to obtain a first target image.

[0060] In one example, the initial target object detection model may be a fusion model, and the illumination enhancement layer in the model may be an SCI illumination enhancement layer. The SCI illumination enhancement layer performs illumination processing on the input image to obtain a first target image. The first target image is an image after illumination processing. For example, the input image is a low-light image, and the first target image is a clear image.

[0061] S104: Identify the first target image based on the target object detection layer in the initial target object detection model to determine actual label information of the target object in the first target image.

[0062] In one example, the target object detection layer may be composed of a YOLOv8 model, wherein the YOLOv8 model is composed of an input module, a backbone module, a Neck module, and an output terminal.

[0063] In one example, the input module uses a data enhancement method to increase the richness of the data set and the robustness of the network. In this embodiment, mosaic enhancement is used to increase the richness of the data and simulate the complex illumination of the telescope site to increase the robustness of the YOLOv8 model. The backbone module is mainly composed of a C2f module and a convolutional layer. Among them, the convolutional layer mainly extracts the spatial information in the first target image through convolution, and performs batch normalization through the BN layer, and finally performs nonlinear activation through the activation function, thereby converting and extracting the features of the first target image. The C2f module is a residual feature learning module of YOLOv8. By branching and cross-layering the feature map output by the CBS module, the feature map is cut and fused at different layers, and the gradient flow of the model is enriched, and a neural network module with stronger feature representation capabilities can be formed. In one example, the Neck module adopts a Pan feature fusion structure, which is first downsampled and then upsampled. There are two cross-layer fusion connections between the two branches of upsampling and downsampling, which can enhance the network's ability to fuse features of objects of different scaling scales, making the network more sensitive to the actual label information of the target object in the first target image.

[0064] S105: Train an initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0065] In an example, loss calculation and target object screening are implemented according to the real label information of the target object and the actual label information of the target object through the output end in the YOLOv8 model, and then the initial target object detection model is trained to obtain the trained target object detection model.

[0066] The disclosed embodiment provides a method for determining a trained target object detection model, the method comprising: obtaining a road image dataset of a vehicle under different road conditions, performing a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; the real label information is used to characterize the position information and category information of each target object; repeatedly performing the following steps until a preset condition is met; inputting the input image into an initial target object detection model, performing illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model to obtain a first target image; recognizing the first target image based on the target object detection layer in the initial target object detection model to determine the actual label information of the target object in the first target image; training the initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model. The technical solution can accurately detect the target object in a low-light scene, thereby avoiding the influence of light brightness.

[0067] Figure 2a The following is a flow chart of a method for determining a trained target object detection model provided by an embodiment of the present disclosure. The present disclosure embodiment is optimized on the basis of the above embodiment, and the present disclosure embodiment can be combined with various optional solutions in one or more of the above embodiments.

[0068] like Figure 2a As shown, the method for determining the trained target object detection model may include the following steps:

[0069] S201, obtaining a road image dataset of a vehicle under different road conditions, obtaining pixel points in each road image, and obtaining first coordinate information of the pixel points.

[0070] In one example, for example, the first coordinate information of pixel point A in the road image is (x, y).

[0071] S202: Perform a rotation transformation on the first coordinate information of the pixel point according to the affine matrix to obtain the second coordinate information of the pixel point.

[0072] In an example, it can be implemented based on the affine matrix in OpenCV. Specifically, it mainly includes two steps:

[0073] First, use the getRotationMatrix2D method to obtain the affine matrix, then pass in the first coordinate information (x, y), and the rotation angle angle; then use the cv2.warpAffine method in opencv to perform affine transformation using the obtained affine matrix. The implementation formula is as follows:

[0074]

[0075] Among them, m 11 , m 12 , m 13 , m 21 , m 22 , m 23 is a linear transformation parameter used to describe the rotation relationship of the road image, wherein m13 and m23 are translation parameters, and (x', y') is the second coordinate information.

[0076] S203: Determine the input image according to the second coordinate information of the pixel point.

[0077] In one example, each target object in each road image has true label information; the true label information is used to represent the location information and category information of each target object.

[0078] In one example, the input image is determined by transforming each pixel in each road image and combining the transformed pixel points according to the corresponding second coordinate information.

[0079] In this embodiment, multiple frames of continuous input images may be disrupted and reordered in time sequence to reconstruct the input image.

[0080] S204, repeat the following steps until the preset conditions are met.

[0081] In an example, this step may refer to the content of step S102.

[0082] S205 , input the input image to the initial target object detection model, and based on the self-calibration module in the illumination enhancement layer, perform self-calibration on the current illumination intensity of the input image to obtain the calibrated illumination intensity.

[0083] In one example, the self-calibration module makes the illumination intensity of each stage converge to the same state. The self-calibration module can display the difference between the illumination intensity of each stage and the illumination intensity of the first stage. Specifically, the self-calibration module can be expressed as:

[0084] ψ(x t ):

[0085] In the above formula, y is the input image, x t is the light intensity at stage t, z t is the clear image at stage t, k θ is a learnable weight module, which is used to transform z t Converge to the same state to reduce the computational complexity, s t For the convergence result, V t Represents the light intensity after calibration at stage t.

[0086] S206 . Based on the illumination estimation module in the illumination enhancement layer, iterate the calibrated illumination intensity to obtain a first target image.

[0087] In one example, according to Retinex theory, the low-light observed image is equal to the dot product of the clear image and the illumination, that is:

[0088]

[0089] Among them, y is the input image, z represents the clear image, that is, the first target image, and x represents the illumination.

[0090] In one example, based on the illumination estimation module in the illumination enhancement layer, iterating the calibrated illumination intensity to obtain the first target image includes:

[0091] Based on the mapping function in the light estimation module, learn the residual between the calibrated light intensity and the target light intensity;

[0092] Based on the illumination estimation module, a first target image is obtained according to the residual and the calibrated illumination intensity.

[0093] In one example, according to the Retinex theory, the first target image is determined by the following formula:

[0094] F(x t ):

[0095] Among them, x 0 Initial value, is the illumination estimation network, where V t represents the illumination intensity after calibration at stage t, u t represents the residual between the illumination intensity and the target definition at the tth stage, and y represents the input image.

[0096] S207: Identify the first target image based on the target object detection layer in the initial target object detection model, and determine actual label information of the target object in the first target image.

[0097] In one example, before identifying the first target image based on the target object detection layer in the initial target object detection model and determining actual label information of the target object in the first target image, the method further includes:

[0098] Based on the target object detection layer in the initial target object detection model, the first target image is processed to obtain a vertical perception feature image and a horizontal perception feature image;

[0099] A processed first target image is determined according to the perception feature image in the vertical direction, the perception feature image in the horizontal direction and the first target image.

[0100] In one example, in order to improve the detection ability of the initial target object detection model for low-light images, the YOLOv8 model is improved by adding the CA attention mechanism in the Neck module. The CA attention mechanism embeds the position information into the channel attention, so that the lightweight network can pay attention to a larger area while avoiding a large amount of computational overhead. The specific structure can be seen in Figure 2b A schematic diagram of the structure of a CA attention mechanism is shown in FIG. Figure 2bAs shown in the figure, Residual is the residual module, XAvgPool and YAvgPool are pooling in the horizontal and vertical directions, conv2d is the convolution operation, BatchNorm is batch normalization, and Sigmoid is the activation function. Specifically, two one-dimensional global pooling operations are used for the first time to aggregate the input features in the vertical and horizontal directions into two independent direction-aware feature maps. Then the two feature maps embedded with specific direction information are encoded into two attention maps, each of which captures the long-range dependency of the input feature map along a spatial direction. Therefore, the position information is stored in the generated attention map, and the two attention maps are then multiplied by the input feature map to enhance the representation ability of the feature map.

[0101] S208. Train an initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0102] In an example, this step may refer to the content of step S105.

[0103] For a clearer explanation, see Figure 2c A schematic diagram of the structure of a trained target object detection model is shown.

[0104] The disclosed embodiment provides a method for determining a trained target object detection model, the method comprising: inputting an input image into an initial target object detection model, and based on a self-calibration module in an illumination enhancement layer, self-calibrating the current illumination intensity of the input image to obtain a calibrated illumination intensity. Based on an illumination estimation module in the illumination enhancement layer, iterating the calibrated illumination intensity to obtain a first target image. Based on the target object detection layer in the initial target object detection model, the first target image is identified to determine the actual label information of the target object in the first target image. Based on the real label information of the target object and the actual label information of the target object, the initial target object detection model is trained to obtain a trained target object detection model. By adopting the present technical solution and introducing a self-calibration module, the purpose of enhancing illumination using simple operations is achieved.

[0105] Figure 3 is a flow chart of a target object detection method provided by an embodiment of the present disclosure, which can be performed by an electronic device. The electronic device can be exemplarily understood as a device such as a mobile phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, etc. Applied to the trained target object detection model, such as Figure 3 As shown, the method provided in this embodiment includes the following steps:

[0106] S301, obtaining a current road image of a vehicle, and inputting the road image into a trained target object detection model.

[0107] In one example, the current road image of the vehicle may be acquired in real time, and the acquired road image is input into a trained target object detection model.

[0108] S302: Determine a target object in a current road image based on the trained target object detection model.

[0109] In one example, a target object in a current road image is identified by a trained target object detection model, and the target object may be a pedestrian, a truck, or a passenger car.

[0110] The disclosed embodiment provides a target object detection method, which includes: obtaining a current road image of a vehicle, inputting the road image into a trained target object detection model, and determining a target object in the current road image based on the trained target object detection model. The technical solution can accurately detect the target object in a low-light scene, thereby avoiding the influence of light brightness.

[0111] Figure 4 1 is a schematic diagram of a structure of a device for determining a trained target object detection model provided by an embodiment of the present disclosure. The device for determining a trained target object detection model can be understood as the above-mentioned electronic device or a part of the functional modules in the above-mentioned electronic device. Figure 4 As shown, the trained target object detection model determination device 40 includes:

[0112] The first acquisition module 401 is used to acquire a road image dataset of a vehicle under different road conditions, and to perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to characterize the location information and category information of each target object.

[0113] Repeat the following steps until the preset conditions are met;

[0114] The processing module 402 is used to input the input image into the initial target object detection model, perform illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model, and obtain a first target image.

[0115] The recognition module 403 is used to recognize the first target image based on the target object detection layer in the initial target object detection model, and determine the actual label information of the target object in the first target image.

[0116] The training module 404 is used to train an initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0117] In one example, the processing module 402 includes:

[0118] A self-calibration submodule, used to perform self-calibration on the current illumination intensity of the input image based on the self-calibration module in the illumination enhancement layer to obtain a calibrated illumination intensity;

[0119] The iterative submodule is used to iterate the calibrated illumination intensity based on the illumination estimation module in the illumination enhancement layer to obtain a first target image.

[0120] In one example, the iterator module is specifically used to:

[0121] Based on the mapping function in the light estimation module, learn the residual between the calibrated light intensity and the target light intensity;

[0122] Based on the illumination estimation module, a first target image is obtained according to the residual and the calibrated illumination intensity.

[0123] In one example, the first acquisition module 401 includes:

[0124] The acquisition submodule is used to acquire the pixel points in each road image and acquire the first coordinate information of the pixel points;

[0125] A transformation submodule, used to perform a rotation transformation on the first coordinate information of the pixel point according to an affine matrix to obtain the second coordinate information of the pixel point;

[0126] The determination submodule is used to determine the input image according to the second coordinate information of the pixel point.

[0127] In one example, the apparatus 40 further includes:

[0128] A first determination module 405 is used to process the first target image based on the target object detection layer in the initial target object detection model to obtain a vertical perception feature image and a horizontal perception feature image;

[0129] The second determination module 406 is used to determine the processed first target image according to the vertical perception feature image, the horizontal perception feature image and the first target image.

[0130] The device provided in this embodiment can execute the method of any of the above embodiments, and its execution method and beneficial effects are similar, which will not be repeated here.

[0131] Figure 5is a schematic diagram of the structure of a target object detection device provided by an embodiment of the present disclosure, which is applied to a trained target object detection model. The device can be understood as the above-mentioned electronic device or a part of the functional modules in the above-mentioned electronic device. Figure 5 As shown, the device 50 comprises:

[0132] The second acquisition module 501 is used to acquire the current road image of the vehicle and input the road image into the trained target object detection model.

[0133] The third determination module 502 is used to determine the target object in the current road image based on the trained target object detection model.

[0134] An embodiment of the present disclosure further provides an electronic device, which includes: a memory, in which a computer program is stored; and a processor, which is used to execute the computer program. When the computer program is executed by the processor, the method of any of the above embodiments can be implemented.

[0135] For example, Figure 6 Schematic diagram of the structure of an electronic device in the embodiment of the present disclosure. Figure 6 , which shows a schematic diagram of the structure of an electronic device 1000 suitable for implementing the embodiment of the present disclosure. The electronic device 1000 in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0136] like Figure 6 As shown, the electronic device 1000 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the electronic device 1000 are also stored. The processing device 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0137] Typically, the following devices may be connected to the I / O interface 1005: an input device 1006 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1007 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1008 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device 1000 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device 1000 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0138] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 1009, or installed from a storage device 1008, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0139] It should be noted that the computer-readable medium disclosed above may 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 be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, 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 disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0140] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0141] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0142] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a road image data set of a vehicle under different road conditions, and performs a rotation transformation on each road image in the road image data set to obtain an input image; wherein each target object in each road image has real label information; the real label information is used to characterize the location information and category information of each target object; repeatedly performs the following steps until a preset condition is met; inputs the input image into an initial target object detection model, and performs illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model to obtain a first target image; recognizes the first target image based on the target object detection layer in the initial target object detection model, and determines the actual label information of the target object in the first target image; trains the initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

[0143] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0144] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0145] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0146] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0147] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 foregoing.

[0148] The embodiments of the present disclosure also provide a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method of any of the above embodiments can be implemented. The execution method and beneficial effects are similar and will not be repeated here.

[0149] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0150] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining a trained target object detection model, characterized in that: The method comprises: Acquire a road image dataset of a vehicle under different road conditions, and perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to represent the position information and category information of each target object; Repeat the following steps until the preset conditions are met; Inputting the input image into an initial target object detection model, performing illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model, to obtain a first target image; Recognize the first target image based on the target object detection layer in the initial target object detection model, and determine actual label information of the target object in the first target image; The initial target object detection model is trained according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

2. The method according to claim 1, characterized in that The step of performing illumination processing on the input image based on the illumination enhancement layer in the initial target object detection model to obtain a first target image includes: Based on the self-calibration module in the illumination enhancement layer, self-calibrate the current illumination intensity of the input image to obtain a calibrated illumination intensity; Based on the illumination estimation module in the illumination enhancement layer, the calibrated illumination intensity is iterated to obtain a first target image.

3. The method according to claim 2, characterized in that The step of iterating the calibrated illumination intensity based on the illumination estimation module in the illumination enhancement layer to obtain a first target image includes: Based on the mapping function in the light estimation module, learning the residual between the calibrated light intensity and the target light intensity; Based on the illumination estimation module, the first target image is obtained according to the residual and the calibrated illumination intensity.

4. The method according to claim 1, characterized in that The step of performing a rotation transformation on each road image in the road image dataset to obtain an input image comprises: Acquire pixel points in each road image, and acquire first coordinate information of the pixel points; According to the affine matrix, the first coordinate information of the pixel point is rotated and transformed to obtain the second coordinate information of the pixel point; The input image is determined according to the second coordinate information of the pixel point.

5. The method according to claim 1, characterized in that Before identifying the first target image based on the target object detection layer in the initial target object detection model and determining actual label information of the target object in the first target image, the method further includes: Based on the target object detection layer in the initial target object detection model, the first target image is processed to obtain a perception feature image in a vertical direction and a perception feature image in a horizontal direction; A processed first target image is determined according to the vertical perception feature image, the horizontal perception feature image and the first target image.

6. A target object detection method, characterized in that: Applied to a trained target object detection model, the method comprises: Obtaining a current road image of the vehicle, and inputting the road image into a trained target object detection model; Based on the trained target object detection model, a target object in the current road image is determined.

7. A device for determining a trained target object detection model, characterized in that: The device comprises: A first acquisition module is used to acquire a road image dataset of a vehicle under different road conditions, and to perform a rotation transformation on each road image in the road image dataset to obtain an input image; wherein each target object in each road image has real label information; and the real label information is used to represent the position information and category information of each target object; Repeat the following steps until the preset conditions are met; A processing module, configured to input the input image into an initial target object detection model, and perform illumination processing on the input image based on an illumination enhancement layer in the initial target object detection model to obtain a first target image; A recognition module, configured to recognize the first target image based on the target object detection layer in the initial target object detection model, and determine actual label information of the target object in the first target image; The training module is used to train the initial target object detection model according to the real label information of the target object and the actual label information of the target object to obtain a trained target object detection model.

8. A target object detection device, characterized in that: Applied to a trained target object detection model, the device comprises: A second acquisition module is used to acquire a current road image of the vehicle and input the road image into a trained target object detection model; The third determination module is used to determine the target object in the current road image based on the trained target object detection model.

9. An electronic device, characterized in that: include: A processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the method of any one of claims 1 to 5 or claim 6.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 or claim 6 is implemented.

Images