Target detection method, device, equipment and storage medium for autonomous driving

Abstract

The present application relates to a target detection method, device, equipment and storage medium for autonomous driving. The method includes: obtaining an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model; downsampling the original image to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; cross-multi-sampling the original image to obtain at least two local images; wherein the resolution of any local image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; inputting the global image and each local image into the preset target detection model to obtain the detection result output by the preset target detection model. The solution provided by the present application can receive and process target detection tasks of high-resolution images and improve the detection rate of small target objects in the image.

Description

Technical Field

[0001] The present application relates to the field of navigation technology, and in particular to a target detection method, device, equipment and storage medium for autonomous driving. Background Art

[0002] Object detection is a traditional task in the field of computer vision. Unlike image recognition, object detection not only needs to identify the objects in the image and give the corresponding categories, but also needs to give the location of the object in the form of a minimum bounding box. Object detection tasks can be completed by object detection models (such as YOLO model, Faster R-CNN model, SSD model, etc.). Object detection models can detect input images to identify the location and category of target objects in the image. In the field of navigation technology, object detection technology has been widely used, providing technical support for the realization of autonomous driving functions of automobiles.

[0003] However, the target detection model cannot process images with too high resolution. Although the resolution of the original image can be reduced to meet the target detection model's upper limit on the input image resolution, it will cause the loss of information about small target objects in the image, affecting the model's recognition detection rate for small target objects. Summary of the invention

[0004] In order to solve or partially solve the problems existing in the related technology, the present application provides a target detection method, device, equipment and storage medium for autonomous driving, which can receive and process target detection tasks of high-resolution images and improve the detection rate of small target objects in the image.

[0005] The first aspect of the present application provides a target detection method for autonomous driving, comprising:

[0006] Acquire an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0007] Downsampling the original image to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0008] The original image is subjected to cross-multi-sampling processing to obtain at least two partial images; wherein the resolution of any of the partial images is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0009] The global image and each of the local images are input into the preset target detection model to obtain a detection result output by the preset target detection model.

[0010] In one embodiment, after inputting the global image and each of the local images into the preset target detection model and obtaining the detection result output by the preset target detection model, the method further includes:

[0011] Generate global feature data based on the intermediate feature data of the preset target detection model;

[0012] Whether global information in the detection result is lost is determined according to the global feature data and the global image.

[0013] In one embodiment, generating global feature data according to the intermediate feature data of the preset target detection model includes:

[0014] Extracting intermediate feature data of the preset target detection model using a generation network in a preset evaluation module pre-connected to the preset target detection model, and generating global feature data based on the intermediate feature data;

[0015] The determining, based on the global feature data and the global image, whether global information in the detection result is lost includes:

[0016] The global feature data and the global image are acquired by using a discriminant network in a preset judging module, and whether the global information in the detection result is lost is judged according to the similarity between the global feature data and the global image.

[0017] In one implementation, after determining whether the global information in the detection result is lost, the method further includes:

[0018] After determining that the global information in the detection result is lost, the cross-multi-sampling processing mode of the original image is adjusted and / or the down-sampling processing mode of the original image is adjusted.

[0019] In one implementation, downsampling the original image to obtain a global image includes:

[0020] The original image is downsampled and border-filled to obtain a global image that meets the size requirements of the preset target detection model for the input image.

[0021] In one implementation, all feature information in the original image is sampled and extracted by each of the local images, and each of the local images represents feature information of a portion of the original image.

[0022] A second aspect of the present application provides a target detection device for autonomous driving, comprising:

[0023] An acquisition module, used to acquire an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0024] A first sampling module is used to downsample the original image acquired by the acquisition module to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0025] A second sampling module is used to perform cross-multi-sampling processing on the original image acquired by the acquisition module to obtain at least two partial images; wherein the resolution of any of the partial images is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model;

[0026] The detection module is used to input the global image obtained by the first sampling module and the local images obtained by each of the second sampling modules into the preset target detection model obtained by the acquisition module to obtain the detection result output by the preset target detection model.

[0027] In one embodiment, the device further comprises:

[0028] A generation module, used to generate global feature data according to the intermediate feature data of the preset target detection model that outputs the detection result in the detection module;

[0029] A judgment module is used to judge whether the global information in the detection result is lost according to the global feature data generated by the generation module and the global image obtained by the first sampling module.

[0030] A third aspect of the present application provides an electronic device, including:

[0031] Processor; and

[0032] The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method as described above.

[0033] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as described above.

[0034] The technical solution provided by this application may have the following beneficial effects:

[0035] The method provided by the present application performs downsampling and cross-multi-sampling processing on the original image whose resolution is higher than the upper limit threshold of the input image resolution of the preset target detection model, thereby obtaining a global image and at least two local images whose resolution is not higher than the upper limit threshold of the input image resolution of the preset target detection model. In this way, the preset target detection model can obtain all the feature information in the original image by receiving the global image and at least two local images, avoid the loss of information of small target objects in the original image, so that the information of small target objects in the original image can be input into the model without loss, so that the model can better learn the global information and local information in the original image, and improve the recognition detection rate of small target objects.

[0036] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0038] Figure 1 is a flow chart of a target detection method for autonomous driving shown in an embodiment of the present application;

[0039] Figure 2 is another flow chart of a target detection method for autonomous driving shown in an embodiment of the present application;

[0040] Figure 3 is a schematic diagram of an original image, a global image, and a local image shown in an embodiment of the present application;

[0041] Figure 4 is a schematic diagram of the network structure of a preset target detection model shown in an embodiment of the present application;

[0042] Figure 5 Schematic diagram of the network structure of the preset target detection model and the preset evaluation module shown in the embodiment of the present application;

[0043] Figure 6 yes Figure 5 A partial area enlarged schematic diagram;

[0044] Figure 7 is a schematic diagram of the structure of a target detection device for autonomous driving shown in an embodiment of the present application;

[0045] Figure 8 is another structural schematic diagram of a target detection device for autonomous driving shown in an embodiment of the present application;

[0046] Fig. 9 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0048] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0049] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0050] In the related art, the target detection model cannot process images with too high resolution. Although the resolution of the original image can be reduced to meet the target detection model's requirement for the upper limit of the input image resolution, it will cause the loss of information of small target objects in the image, affecting the model's recognition detection rate for small target objects.

[0051] In response to the above problems, an embodiment of the present application provides a target detection method for autonomous driving, which is capable of receiving and processing target detection tasks of high-resolution images and improving the detection rate of small target objects in the image.

[0052] The technical solution of the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0053] Figure 1 It is a flow chart of a target detection method for autonomous driving shown in an embodiment of the present application.

[0054] See also Figure 1, the method comprising:

[0055] Step S101, obtaining an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model.

[0056] Among them, the preset target detection model can be a One-stage target detection model (such as YOLO, SSD, SqueezeDet, DetectNet) or a Two-stage target detection model (such as Faster R-CNN).

[0057] The original image may be an image captured by a camera mounted on the vehicle while the vehicle is driving. The resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model. For example, the preset target detection model is the yolo-v5m model, and the threshold of the upper limit of the input image resolution of the model is 640*640. If the original image is 1920*1080, the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model.

[0058] Step S102, downsampling the original image to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution set by the preset target detection model.

[0059] In this step, the resize function can be used to downsample the original image, thereby reducing the resolution of the original image to obtain a global image. The resolution of the global image obtained after processing should not be higher than the threshold of the upper limit of the input image resolution of the preset target detection model, that is, the resolution of the global image is less than or equal to the threshold of the upper limit of the input image resolution of the preset target detection model. In this way, the preset target detection model can receive and process the global image and perform target detection tasks on the global image.

[0060] Step S103, cross-multi-sampling the original image to obtain at least two partial images; wherein the resolution of any partial image is not higher than the threshold of the upper limit of the input image resolution set by the preset target detection model.

[0061] In one optional embodiment, performing cross-multi-sampling processing on the original image to obtain at least two local images may include: performing cross-overlapping sampling processing on the original image to obtain at least two local images. Further, all feature information in the original image is sampled and extracted by each local image, and each local image represents the feature information of a part of the original image. In other words, all feature information in the original image can be obtained by obtaining feature information from each local image. Further, the resolution of each local image obtained can be the threshold of the upper limit of the input image resolution of the preset target detection model. Further, the size of each local image may be the same or different, and the feature information of two adjacent local images may be completely different or partially the same.

[0062] It should be noted that cross-overlapping sampling of the original image can be understood as partially intercepting the original image to obtain a local image, and the feature information in the obtained local image is consistent with the feature information of the corresponding area in the original image without loss.

[0063] Step S104: input the global image and each local image into a preset target detection model to obtain a detection result output by the preset target detection model.

[0064] In this step, the preset target detection model can output the detection results of the global image and each local image according to the received global image and each local image. The detection results may include the location information and category information of multiple different target objects (including large target objects and small target objects).

[0065] It can be seen from this embodiment that by downsampling and cross-multi-sampling the original image whose resolution is higher than the upper threshold of the input image resolution of the preset target detection model, a global image and at least two local images whose resolution is not higher than the upper threshold of the input image resolution of the preset target detection model are obtained. In this way, the preset target detection model can obtain all the feature information in the original image by receiving the global image and at least two local images, avoid the loss of information of small target objects in the original image, so that the information of small target objects in the original image can be input into the model without loss, so that the model can better learn the global information and local information in the original image, and improve the recognition detection rate of small target objects.

[0066] Figure 2 It is another flowchart of the method for generating lane lines in an embodiment of the present application. Figure 2 relatively Figure 1 The scheme of the present application is described in more detail.

[0067] See also Figure 2 , the method comprising:

[0068] Step S201, obtaining an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model.

[0069] For this step, please refer to the description in step S101 and will not be repeated here.

[0070] Step S202: downsample the original image to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution set by the preset target detection model.

[0071] For this step, please refer to the description of step S102.

[0072] Furthermore, in an optional implementation of this step, the original image may be downsampled and the borders may be filled to obtain a global image that meets the size requirements of the preset target detection model for the input image. The border filling padding method may include adding pads to the insufficient parts of the image, such as adding black elements to fill the insufficient parts of the image.

[0073] For example, if the preset target detection model requires the input image size to be 640*640, and the original image 1920*1080 is downsampled to 480*640, then padding can be added to the upper and lower boundaries of the 480*640 image to make the image size 640*640, thus meeting the preset target detection model's input image size requirement. For details, please refer to Figure 3 In one embodiment, the original image is Figure 3 In the middle (a), the global image is Figure 3 In the middle (b), it can be seen that the global image is obtained by downsampling the original image and then filling the borders, where the border filling method is to add a black pad. When the preset target detection model performs the target detection task, the black part of the border filling can be regarded as the background, so that the detection accuracy of the preset target detection model is not affected.

[0074] It is understandable that Figure 3 As shown, for a large target object in the original image, such as a 1920*1080 original image ( Figure 3 The car in the middle (a) is downsampled by 3 times to obtain a 640*640 global image ( Figure 3 (b), it can be seen that the information loss of the car in the image is negligible.

[0075] Step S203, cross-multi-sampling the original image to obtain at least two partial images; wherein the resolution of any partial image is not higher than the threshold of the upper limit of the input image resolution set by the preset target detection model.

[0076] For this step, please refer to the description of step S103.

[0077] For further information, please also refer to Figure 3 In one embodiment, the original image is Figure 3 In the middle (a), the global image is Figure 3 In the middle (b), the local image is Figure 3 It can be seen that in this embodiment, the eight images shown in Figure (c) are Figure 3 The original image shown in Figure (a) is processed by cross multi-sampling to obtain the following Figure 3 The eight partial images shown in Figure (c) are shown in Figure 1. Specifically, the resolution of the original image can be 1920*1080, and the threshold of the upper limit of the input image resolution of the model is 640*640, so the resolution of each partial image can be 640*640.

[0078] It is understandable that Figure 3 As shown, for small target objects in the original image, such as the original image of 1920*1080 ( Figure 3 The traffic light in the middle (a) is downsampled by 3 times to obtain a 640*640 global image ( Figure 3 It can be seen that the information of the traffic light in the image is largely lost, and after cross multi-sampling processing, eight local images of 640*640 are obtained ( Figure 3 (c) The information of the traffic light in the image will not be lost.

[0079] Step S204: input the global image and each local image into a preset target detection model to obtain a detection result output by the preset target detection model.

[0080] For this step, please refer to the description in step S104 and will not be repeated here.

[0081] It should be pointed out that if only the global image is used as the input image of the preset target detection model, and the global image is obtained by downsampling the original image, the downsampling process of the original image is likely to cause information loss of small target objects, making the global image less available information, which will in turn reduce the detection recall ability of the model and affect the model's detection rate of small target objects in the image.

[0082] It should also be noted that the detection result output by the preset target detection model in step S204 can effectively ensure a good detection rate for small target objects in the image without affecting the detection rate of large targets. However, when the proportion of local images as input images of the preset target detection model is too large (for example, the global image is 1, the local images are 8, and the proportion of local images is 8 / 9), it is not ruled out that the model may cause loss of expression of global information in the original image (for example, the positioning accuracy of large target objects in the image is reduced, etc.). Accordingly, the embodiment of the present application can continue to perform the following steps to judge whether the global information in the detection result obtained in step S204 is lost, so as to ensure that the detection result obtained in step S204 is correct and reliable.

[0083] Step S205: Generate global feature data based on the intermediate feature data of the preset target detection model.

[0084] In an optional implementation, generating global feature data according to the intermediate feature data of the preset target detection model includes:

[0085] The intermediate feature data of the preset target detection model is extracted by using the generation network in the preset evaluation module pre-connected to the preset target detection model, and the global feature data is generated according to the intermediate feature data.

[0086] The preset evaluation module may be pre-built and connected to the preset target detection model, and the preset evaluation module includes a generation network. The generation network may extract intermediate feature data from the preset target detection model and generate global feature data based on the extracted intermediate feature data.

[0087] Step S206: Determine whether the global information in the detection result is lost based on the global feature data and the global image.

[0088] In an optional implementation, judging whether global information in the detection result is lost according to the global feature data and the global image includes:

[0089] The discriminant network in the preset evaluation module is used to obtain global feature data and global images, and based on the similarity between the global feature data and the global image, it is determined whether the global information in the detection result is lost.

[0090] Among them, the preset evaluation module also includes a discriminant network, which can obtain the global image and the global feature data output by the generation network, so as to judge whether the global information in the detection result is lost according to the similarity between the global feature data and the global image.

[0091] With respect to step S205 and step S206, it should be particularly noted that, in the embodiment of the present application, in order to judge whether the global information in the detection result output in step S204 is lost, the global feature data generated by the intermediate feature data is used to characterize the global information in the detection result, and then the degree of similarity between the global feature data and the global image is judged to determine whether the global information in the detection result is lost.

[0092] In one embodiment, please refer to Figures 4 to 6 The preset target detection model is the yolo-v5 model, and its network structure is as follows Figure 4 The network structure of the preset evaluation module can be found in Figure 5 and Figure 6 , Figure 5 Relative to Figure 4 The added network structure is the preset evaluation module pre-built and connected to the preset target detection model in the embodiment of the present application. In this embodiment, the preset evaluation module includes a generation network and a discrimination network.

[0093] The generation network includes: a first Deconv module, a Concat module, three intermediate Deconv modules, a Conv2d module, and a Goutput module. The generation network uses the first Deconv module to upsample the intermediate feature data extracted from one of the Upsamplingconcat modules of the yolo-v5 model. The Concat module merges the intermediate feature data of the Upsampling concat module of the latter level in the yolo-v5 model with the intermediate feature data after upsampling processing by the first Deconv module. The output data of the Concat module is deconvolved by the three intermediate Deconv modules and convolved by the Conv2d module, and then the generated global feature data is output through the Goutput module.

[0094] Among them, the discriminant network includes: an SPP module, two Conv2d modules, an FC module, and a Doutput module. Among them, the function of the SPP module is to fuse multiple receptive fields, which can be used to determine whether two input data are familiar. The discriminant network uses the SPP module to receive the global feature data output by the Goutput module of the generation network and the input0 global image. After convolution by two Conv2d modules and full connection of the FC module, the judgment signal is output through the Doutput module. In this embodiment, the Doutput module outputs 0 or 1 to indicate whether the global information in the detection result is lost. When the global information in the detection result is lost, it means that the generated global feature data is false, and the supplement of the local image causes the global information in the detection result of the preset target detection model to be lost. At this time, the Doutput module outputs 0. When the global information in the detection result is not lost, it means that the generated global feature data is true, that is, the intermediate feature data can completely restore the global information in the global image, and the supplement of the local image does not cause the loss of global information in the detection result of the preset target detection model. At this time, the Doutput module outputs 1.

[0095] Step S207: after determining that the global information in the detection result is lost, adjusting the cross multi-sampling processing mode of the original image and / or adjusting the down-sampling processing mode of the original image.

[0096] In this step, after determining that the global information in the detection result is lost, the cross-multi-sampling processing method of the original image can be adjusted, the down-sampling processing method of the original image can be adjusted, or the cross-multi-sampling processing and down-sampling processing methods of the original image can be adjusted at the same time. In this way, the data of the detection result output by the preset target detection model can be adjusted so that the global information in the detection result output by the preset target detection model can be achieved without loss, thereby improving the accuracy and reliability of the detection result.

[0097] The method of adjusting the original image for downsampling includes but is not limited to adjusting the ratio of downsampling of the original image and the proportion of border filling. The method of adjusting the original image for cross-multi-sampling includes but is not limited to adjusting the ratio of cross-multi-sampling of the original image and the number of cross-multi-sampling (i.e., the number of local images).

[0098] It can be seen from this embodiment that the global feature data can be generated using the intermediate feature data of the preset target detection model, and based on the global feature data and the global image, it is determined whether the global information in the detection result of the preset target detection model is lost. In this way, the detection result output by the preset target detection model can be further verified, that is, the supplementation of the local image (i.e., local information) does not lead to the loss of global information, thereby ensuring the correctness and reliability of the detection result.

[0099] Corresponding to the aforementioned application function implementation method embodiment, the present application also provides a target detection device, electronic device and corresponding embodiments for autonomous driving.

[0100] Figure 7 It is a schematic diagram of the structure of a target detection device for autonomous driving shown in an embodiment of the present application.

[0101] See also Figure 7 The device includes: an acquisition module 701, a first sampling module 702, a second sampling module 703, and a detection module 704.

[0102] The acquisition module 701 is used to acquire the original image and the preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model.

[0103] The first sampling module 702 is used to downsample the original image acquired by the acquisition module 701 to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution set by the preset target detection model.

[0104] The second sampling module 703 is used to perform cross-multi-sampling processing on the original image acquired by the acquisition module 701 to obtain at least two local images; wherein the resolution of any local image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model.

[0105] The detection module 704 is used to input the global image obtained by the first sampling module 702 and the local images obtained by each second sampling module 703 into the preset target detection model obtained by the acquisition module 701 to obtain the detection result output by the preset target detection model.

[0106] It can be seen from this embodiment that the target detection device for autonomous driving provided in the embodiment of the present application can obtain all feature information in the original image by receiving a global image and at least two local images, thereby avoiding the loss of information of small target objects in the original image, so that the information of small target objects in the original image can be input into the model without loss, so that the model can better learn the global information and local information in the original image, thereby improving the recognition detection rate of small target objects.

[0107] See also Figure 8 , is another structural diagram of a target detection device for autonomous driving shown in an embodiment of the present application. Further, Figure 8 The device shown may also include: an acquisition module 701 , a first sampling module 702 , a second sampling module 703 , a detection module 704 , a generation module 705 , a judgment module 706 , and an adjustment module 707 .

[0108] The functions of the acquisition module 701, the first sampling module 702, the second sampling module 703, and the detection module 704 can be found in Figure 7 The description in the embodiments will not be repeated here.

[0109] The generation module 705 is used to generate global feature data according to the intermediate feature data of the preset target detection model that outputs the detection result in the detection module 704.

[0110] The judgment module 706 is used to judge whether the global information in the detection result is lost according to the global feature data generated by the generation module 705 and the global image obtained by the first sampling module 702.

[0111] The adjustment module 707 is used to adjust the cross-multi-sampling processing mode of the original image and / or adjust the down-sampling processing mode of the original image after the judgment module 706 determines that the global information in the detection result is lost.

[0112] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0113] Fig. 9 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application.

[0114] See also Fig. 9 , the electronic device 900 includes a memory 910 and a processor 920 .

[0115] The processor 920 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0116] The memory 910 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, ROM can store static data or instructions required by the processor 920 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory may store some or all instructions and data required by the processor at run time. In addition, the memory 910 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 910 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and transient electronic signals transmitted wirelessly or wired.

[0117] The memory 910 stores executable codes, and when the executable codes are processed by the processor 920 , the processor 920 can execute part or all of the methods described above.

[0118] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0119] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0120] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A target detection method for autonomous driving, characterized in that: include: Acquire an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model; Downsampling the original image to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; The original image is subjected to cross-multi-sampling processing to obtain at least two partial images; wherein the resolution of any of the partial images is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; Inputting the global image and each of the local images into the preset target detection model to obtain a detection result output by the preset target detection model; Generate global feature data based on the intermediate feature data of the preset target detection model; Determining whether global information in the detection result is lost according to the global feature data and the global image; After determining that the global information in the detection result is lost, the cross-multi-sampling processing mode of the original image is adjusted and / or the down-sampling processing mode of the original image is adjusted.

2. The method according to claim 1, characterized in that The generating global feature data according to the intermediate feature data of the preset target detection model includes: Extracting intermediate feature data of the preset target detection model using a generation network in a preset evaluation module pre-connected to the preset target detection model, and generating global feature data based on the intermediate feature data; The determining, based on the global feature data and the global image, whether global information in the detection result is lost includes: The global feature data and the global image are acquired by using a discriminant network in a preset judging module, and whether the global information in the detection result is lost is judged according to the similarity between the global feature data and the global image.

3. The method according to any one of claims 1 to 2, characterized in that: The downsampling process of the original image to obtain a global image includes: The original image is downsampled and border-filled to obtain a global image that meets the size requirements of the preset target detection model for the input image.

4. The method according to any one of claims 1-2, characterized in that: All feature information in the original image is sampled and extracted by each of the local images, and each of the local images represents feature information of a portion of the original image.

5. A target detection device for autonomous driving, characterized in that: include: An acquisition module, used to acquire an original image and a preset target detection model; wherein the resolution of the original image is higher than the threshold of the upper limit of the input image resolution of the preset target detection model; A first sampling module is used to downsample the original image acquired by the acquisition module to obtain a global image; wherein the resolution of the global image is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; A second sampling module is used to perform cross-multi-sampling processing on the original image acquired by the acquisition module to obtain at least two partial images; wherein the resolution of any of the partial images is not higher than the threshold of the upper limit of the input image resolution of the preset target detection model; A detection module, used to input the global image obtained by the first sampling module and the local images obtained by each of the second sampling modules into the preset target detection model obtained by the acquisition module, to obtain a detection result output by the preset target detection model; A generating module, used to generate global feature data according to the intermediate feature data of the preset target detection model that outputs the detection result in the detection module; A judging module, configured to judge whether global information in the detection result is lost according to the global feature data generated by the generating module and the global image obtained by the first sampling module; The adjustment module is used to adjust the cross-multi-sampling processing mode of the original image and / or adjust the down-sampling processing mode of the original image after the judgment module determines that the global information in the detection result is lost.

6. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having executable codes stored thereon, which, when executed by a processor of an electronic device, causes the processor to execute the method according to any one of claims 1 to 4.

Images