Target detection method and apparatus
By introducing an area prediction model into the target detection model and using regression lines to fit and predict the target area, the problem of insufficient target detection accuracy in autonomous driving is solved, achieving higher detection accuracy and reliability of autonomous driving decisions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JILUO TECH (SHANGHAI) CO LTD
- Filing Date
- 2021-11-30
- Publication Date
- 2026-05-08
AI Technical Summary
Existing object detection models have poor accuracy in autonomous driving, resulting in errors in the detection results.
The target detection results are feature extracted and predicted by the area prediction model. The predicted area of the target is obtained by fitting a regression line and compared with the actual area to judge the accuracy of the detection results.
This improves the accuracy of target detection. By comparing the deviation between the predicted area and the actual area, the accuracy of the detection results is determined, thereby enhancing the decision-making accuracy of autonomous driving.
Smart Images
Figure CN114399546B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and in particular to a target detection method and apparatus. Background Technology
[0002] Object detection is a crucial research area in computer vision. In recent years, deep learning and neural network technologies have made continuous breakthroughs in object detection tasks. In the field of autonomous driving, with the development and upgrading of autonomous driving technology, visual perception solutions based on deep learning and neural network technologies are gradually being widely applied. Autonomous driving often requires avoiding targets such as pedestrians and vehicles; therefore, object detection is critical, and the accuracy of object detection determines the autonomous driving decisions.
[0003] Currently, the most common accuracy testing methods for object detection models involve testing the accuracy of the trained object detection model before putting it into use, and then putting it into use only when the detection accuracy reaches a certain threshold, in order to improve the accuracy of object detection.
[0004] Because the accuracy of the models that have passed the accuracy test and are put into use is fixed and cannot reach 100% accuracy, there are still some errors in the detection results output by the models. Summary of the Invention
[0005] This invention provides a target detection method and apparatus to address the shortcomings of poor target detection accuracy in the prior art, thereby improving the accuracy of target detection results.
[0006] This invention provides a target detection method, comprising: performing target detection on an acquired target image to obtain a target detection result; inputting the target detection result into an area prediction model to obtain a target prediction area output by the area prediction model; wherein the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground truth value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result; obtaining the actual target area based on the target detection result; and determining whether the target detection result is accurate based on the target prediction area and the actual target area to obtain an accuracy detection result.
[0007] According to a target detection method provided by the present invention, the area prediction model includes: a feature extraction layer, which extracts features from the input target detection result to obtain the center point position coordinates; and a target area prediction layer, which predicts the target area based on the center point position coordinates.
[0008] According to a target detection method provided by the present invention, the step of predicting the target area based on the center point position coordinates includes: predicting the target area based on the center point position coordinates using a pre-fitted regression line to obtain the target predicted area.
[0009] According to a target detection method provided by the present invention, before performing feature extraction on the input target detection result, the method further includes: performing edge processing on the target image to be detected.
[0010] According to a target detection method provided by the present invention, the step of determining whether the target detection result is accurate based on the predicted target area and the actual target area to obtain an accuracy detection result includes: obtaining a deviation value based on the predicted target area and the pre-acquired actual target area; and comparing the deviation value with a preset range to obtain an accuracy detection result.
[0011] According to a target detection method provided by the present invention, training the area prediction model includes: acquiring a target training image and its corresponding target area ground truth; performing target detection on the target training image to obtain a target prediction result; using the target prediction result as input data for training and the target area ground truth as a label to train the model to be trained, thereby obtaining an area prediction model for generating the target area prediction result.
[0012] The present invention also provides a target detection device, comprising: a target detection module for performing target detection on an acquired target image to obtain a target detection result; an area prediction module for inputting the target detection result into an area prediction model to obtain a target prediction area output by the area prediction model; wherein the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground truth value corresponding to the target training image; the area prediction model is used to predict the target prediction area based on the position coordinates extracted from the target detection result; an area acquisition module for obtaining the actual area of the target based on the target detection result; and an accuracy detection module for determining whether the target detection result is accurate based on the target prediction area and the actual area of the target to obtain an accuracy detection result.
[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the target detection methods described above.
[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the target detection methods described above.
[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the target detection methods described above.
[0016] The target detection method and apparatus provided by the present invention predicts the area based on the target detection results, so as to compare the predicted area of the target with the actual area of the target obtained based on the target detection results, thereby determining the accuracy of the target detection results and further improving the accuracy of target detection. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating the target detection method provided by the present invention;
[0019] Figure 2 This is a flowchart illustrating the training area prediction model provided by the present invention;
[0020] Figure 3 This is a schematic diagram of the target detection device provided by the present invention;
[0021] Figure 4 This is a schematic diagram of the training module provided by the present invention;
[0022] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0024] Figure 1 A flowchart of a target detection method according to the present invention is shown. The method includes:
[0025] S11, Perform target detection on the acquired target image to obtain the target detection result;
[0026] S12, Input the target detection result into the area prediction model to obtain the target prediction area output by the area prediction model; wherein, the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result;
[0027] S13, Based on the target detection results, obtain the actual area of the target;
[0028] S14. Based on the predicted area and the actual area of the target, determine whether the target detection result is accurate, and obtain the accuracy detection result.
[0029] It should be noted that S1N in this specification does not represent the order of the target detection methods. The following details will explain this in conjunction with... Figure 2 The target detection method of the present invention is described.
[0030] Step S11: Perform target detection on the acquired target image to obtain the target detection result.
[0031] In this embodiment, target detection is performed on the acquired target image to be detected to obtain the target detection result, including: inputting the acquired target image to be detected into the target detection model to obtain the target detection result output by the target detection model, wherein the target detection model is trained based on the target training samples and their corresponding target detection ground values.
[0032] In one optional embodiment, target detection is performed on the acquired target image to obtain target detection results, including: performing target detection on the acquired target image to obtain target detection results; and labeling the detected targets and their corresponding positions to obtain target detection results.
[0033] In an optional embodiment, before performing target detection on the acquired target image, the method further includes: obtaining the target image to be detected based on an electronic device or application platform using the target detection method; or, obtaining the target image based on a terminal device connected to the electronic device or application platform using the target detection method. It should be noted that the aforementioned terminal device or application platform can obtain the target image through a connected visual sensor, including millimeter-wave radar, lidar, detectors, cameras, and other imaging devices.
[0034] It should be noted that the acquired target image can be a single frame or a sequence of image frames captured from a target in a linear scene, or a video of a target in a linear scene that has undergone shot segmentation to obtain image frames or a sequence of image frames. Furthermore, the target can be a pedestrian or a vehicle, and a linear scene is a scene where the target's position and size in the image do not change drastically.
[0035] Step S12: Input the target detection result into the area prediction model to obtain the target prediction area output by the area prediction model; wherein, the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result.
[0036] In this embodiment, the area prediction model includes: a feature extraction layer, which extracts features from the input target detection result to obtain the center point position coordinates; and a target area prediction layer, which predicts the target area based on the center point position coordinates.
[0037] Specifically, firstly, the feature extraction layer extracts features from the input target detection results to obtain the center point coordinates. In this embodiment, extracting features from the input target detection results to obtain the center point coordinates includes: extracting target contour features based on the target detection results; and obtaining the center point coordinates based on the extracted contour features. It should be noted that after obtaining the target detection results, the center point coordinates are determined based on the extracted features to facilitate subsequent prediction of the target's area based on the center point coordinates. This allows for determining the accuracy of the target detection results by comparing the predicted target area with the actual target area in the scene, thereby identifying any false detections.
[0038] In an optional embodiment, before feature extraction from the input target detection result, the method further includes: edge processing of the target image to be detected. Specifically, this includes: adding edge pixels to the target image to be detected based on preset pixel values, where the preset pixel values can be 0 or RGB color, to ensure that the edges of the image are processed, and removing the edges with added edge pixels after convolution processing.
[0039] Secondly, the target area prediction layer predicts the target area based on the center point's coordinates. More specifically, predicting the target area based on the center point's coordinates includes: using a pre-fitted regression line to predict the target area based on the center point's coordinates. In this embodiment, the regression line is represented as:
[0040]
[0041] in, Let (x,y) represent the target predicted area, (x,y) represent the coordinates of the center point, and k1, k2, and b represent the weight coefficients of the regression line obtained through training.
[0042] In one optional embodiment, the target detection result is input into the area prediction model to obtain the accuracy detection result output by the area prediction model, including: inputting the target detection result into the feature extraction layer to obtain the center point position coordinates output by the feature extraction layer; inputting the center point position coordinates into the target area prediction layer to obtain the target predicted area output by the target area prediction layer; and inputting the target predicted area into the accuracy detection layer to obtain the accuracy detection result output by the accuracy detection layer.
[0043] Step S13: Based on the target detection results, obtain the actual area of the target.
[0044] In this embodiment, obtaining the actual area of the target based on the target detection results includes: obtaining the coordinate position of the target based on the target detection results; and calculating the actual area of the target based on the coordinate position of the target.
[0045] In one optional embodiment, obtaining the actual area of the target based on the target detection result includes: selecting the actual area of the target corresponding to the target detection result from the pre-acquired actual areas of the target corresponding to different targets based on the target detection result.
[0046] Step S14: Determine whether the target detection result is accurate based on the predicted target area and the actual target area, and obtain the accuracy detection result.
[0047] In this embodiment, the accuracy of the target detection result is determined based on the predicted target area and the actual target area to obtain the accuracy detection result. This includes: obtaining a deviation value based on the predicted target area and the pre-acquired actual target area; and comparing the deviation value with a preset range to obtain the accuracy detection result.
[0048] Furthermore, the deviation value is compared with a preset range to obtain the accuracy detection result, including: if the deviation value meets the preset range, the target detection result is determined to be accurate, and this is used as the accuracy detection result; otherwise, the target detection result is determined to be a false detection, and this is used as the accuracy detection result. It should be noted that the preset range can be determined based on the allowable error between the predicted area and the actual area of the target. For example, if the allowable error is ΔS, then the preset range is [-ΔS, ΔS]. If the actual area of the target is S... 实 Then when the target predicted area is in [S 实 -ΔS,S 实When the target detection result is within the range of +ΔS, the corresponding target detection result is accurate, that is, the accuracy detection result is accurate. Otherwise, the accuracy detection result is a false detection.
[0049] In one alternative embodiment, reference Figure 2 The method further includes: training the area prediction model before inputting the target detection results into the area prediction model, specifically including:
[0050] S21, Obtain the target training image and its corresponding target area ground value;
[0051] S22, Perform target detection on the target training image to obtain the target prediction result;
[0052] S23, using the target prediction results as input data for training and the target area ground value as a label, train the model to be trained to obtain an area prediction model for generating target detection results.
[0053] It should be noted that S2N in this specification does not represent the order of the target detection methods.
[0054] Step S21: Obtain the target training image and its corresponding ground truth area.
[0055] In this embodiment, obtaining the target training image and its corresponding target area ground truth includes: obtaining the target training image; annotating the target training image to obtain target annotation data; and obtaining the target area ground truth based on the target annotation data.
[0056] In one optional embodiment, acquiring the target training image includes: acquiring a video stream; capturing a certain number of video frame images at preset intervals as target training images; or, continuously capturing at least one frame image based on at least one target as target training images. It should be noted that when acquiring the video stream or capturing the target image, acquisition can be based on different external factors such as pose angles, obstructions, and lighting.
[0057] In addition to acquiring the target training images, the process also includes: performing target detection on the acquired target training images to remove those that do not contain the target. It should be noted that target training images can be understood as a collection of image data. This image data can be images targeting at least one target, corresponding to each target at different angles, with different pixel colors, etc. The amount of image data is usually quite large, reaching millions. These target training images are used to train the network to be trained, thus completing the model construction.
[0058] Step S22: Perform target detection on the target training image to obtain the target prediction result.
[0059] In this embodiment, target detection on the target training image includes: inputting the acquired target training image into the target detection model to obtain the target prediction result output by the target detection model, wherein the target detection model is trained based on the target training samples and their corresponding target detection ground values; or, performing target detection on the acquired target training image; labeling the detected targets and their corresponding positions to obtain the target prediction result.
[0060] Step S23: Using the target prediction results as input data for training and the target area ground truth as labels, train the model to be trained to obtain an area prediction model for generating target detection results.
[0061] In this embodiment, the network to be trained typically includes a feature extraction layer for extracting the center point position based on the target prediction result, a target area prediction layer for predicting the target area corresponding to the target training image based on the center point position, and a loss function. According to a preset iteration rule, the above sample images or data-enhanced sample images are input into the model to be trained for training to obtain the trained target recognition model.
[0062] Specifically, training the model to be trained includes: inputting the target training image into the feature extraction layer to obtain the center point position output by the feature extraction layer; inputting the center point position into the target area prediction layer to obtain the predicted target area output by the target area prediction layer; constructing a loss function based on the predicted target area and the ground truth of the target area; and converging based on the loss function to end the training.
[0063] It should be noted that the loss function can use the mean squared error (MSE) as a confidence reference for the predicted target area. In this embodiment, the mean squared error is expressed as:
[0064]
[0065] Where MSE represents the mean squared error, and n represents the number of target prediction results. y represents the predicted target area corresponding to the prediction result of the i-th target; i This represents the true value of the target area corresponding to the predicted result of the i-th target.
[0066] In summary, the embodiments of the present invention predict the area based on the target detection results, so as to compare the predicted area of the target with the actual area of the target obtained based on the target detection results, thereby determining the accuracy of the target detection results and further improving the accuracy of target detection.
[0067] The target detection device provided by the present invention is described below. The target detection device described below and the target detection method described above can be referred to in correspondence.
[0068] Figure 3 A schematic diagram of a target detection device is shown, the device comprising:
[0069] The target detection module 31 performs target detection on the acquired target image and obtains the target detection result;
[0070] The area prediction module 32 inputs the target detection result into the area prediction model to obtain the target prediction area output by the area prediction model; wherein, the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result.
[0071] The area acquisition module 33 obtains the actual area of the target based on the target detection results;
[0072] The accuracy detection module 34 determines the accuracy of the target detection result based on the predicted target area and the actual target area, and obtains the accuracy detection result.
[0073] In this embodiment, the target detection module 31 includes: a data transmission unit, which inputs the acquired target image to be detected into the target detection model; a detection unit, in which the target detection model performs target detection on the target image to be detected and obtains the target detection result; and a data output unit, which outputs the target detection result. It should be noted that the target detection model is pre-trained based on target training samples and their corresponding ground truth values.
[0074] In an optional embodiment, the target detection module 31 includes: a detection unit for performing target detection on the acquired target image; and a labeling unit for labeling the detected targets and their corresponding positions to obtain target detection results.
[0075] In an optional embodiment, the device further includes: an image acquisition module, which acquires an image of the target to be detected based on an electronic device or application platform using the target detection method; or, it acquires the image of the target to be detected based on a terminal device connected to the electronic device or application platform using the target detection method. It should be noted that the aforementioned terminal device or application platform can acquire the image of the target to be detected through a visual sensor connected to it, including millimeter-wave radar, lidar, detectors, cameras, and other imaging devices.
[0076] It should be noted that the target image acquired by the image acquisition module can be a single frame or a sequence of image frames captured for a target in a linear scene, or a video of a target in a linear scene that has been segmented to obtain image frames or a sequence of image frames. Furthermore, the target can be a pedestrian or a vehicle, etc. A linear scene refers to a scene where the size of the target at a fixed position in the image does not change drastically.
[0077] The area prediction module 32 includes: a feature extraction unit, which extracts features from the input target detection results to obtain the center point position coordinates; and a target area prediction unit, which predicts the target area based on the center point position coordinates.
[0078] Specifically, the feature extraction unit includes: a feature extraction subunit, which extracts target contour features based on the target detection results; and a coordinate acquisition subunit, which obtains the center point coordinates based on the extracted contour features. It should be noted that after obtaining the target detection results, the center point coordinates are determined based on the extracted features to facilitate subsequent prediction of the target's area based on the center point coordinates. This allows for comparison between the predicted target area and the actual target area in the scene to determine the accuracy of the target detection results and identify any false detections.
[0079] In an optional embodiment, the apparatus further includes an edge processing module for performing edge processing on the target image to be detected. Specifically, the edge processing module includes: a pixel filling unit for adding edge pixels to the target image to be detected based on preset pixel values, where the preset pixel values can be 0 or RGB color, to ensure that the edges of the image are processed; and a pixel processing unit for removing the edges with added edge pixels after convolution processing.
[0080] The target area prediction unit includes: a target area prediction subunit, which predicts the target area based on the center point position coordinates and using a pre-fitted regression line to obtain the target predicted area.
[0081] The area acquisition module 33 includes: a target acquisition unit, which obtains the coordinate position of the target based on the target detection results; and an area acquisition unit, which calculates the actual area of the target based on the coordinate position of the target.
[0082] In an optional embodiment, the area acquisition module 33 includes: an area acquisition unit, which selects the actual area of the target corresponding to the target detection result from the pre-acquired actual areas of the target corresponding to different targets based on the target detection result.
[0083] The accuracy detection module 34 includes: a deviation acquisition unit, which obtains a deviation value based on the predicted target area and the pre-acquired actual target area; and a comparison unit, which compares the deviation value with a preset range to obtain an accuracy detection result.
[0084] Furthermore, in the comparison unit, the deviation value is compared with a preset range. When obtaining the accuracy detection result, if the deviation value meets the preset range, the target detection result is determined to be accurate, and this is used as the accuracy detection result; otherwise, the target detection result is determined to be a false detection, and this is used as the accuracy detection result. It should be noted that the preset range can be determined based on the allowable error between the predicted area and the actual area.
[0085] In one alternative embodiment, reference Figure 4 The device also includes a training module for training the area prediction model before inputting the target detection results into the area prediction model. Specifically, the training module includes:
[0086] Data acquisition unit 41 acquires the target training image and its corresponding target area ground truth value;
[0087] The target detection unit 42 performs target detection on the target training image and obtains the target prediction result;
[0088] Training unit 43 uses the target prediction results as input data for training and the target area ground value as a label to train the model to be trained, thereby obtaining an area prediction model for generating target detection results.
[0089] In this embodiment, the data acquisition unit 41 includes: an image acquisition subunit for acquiring a target training image; an annotation subunit for annotating the target training image to obtain target annotation data; and a truth value acquisition subunit for obtaining the target area truth value based on the target annotation data.
[0090] In one optional embodiment, the image acquisition subunit includes: a video acquisition grandchild unit for acquiring a video stream; and an image acquisition grandchild unit for acquiring a certain number of video frame images at a preset interval as target training images; or, the image acquisition grandchild unit for continuously capturing at least one frame image based on at least one target as target training images. It should be noted that when acquiring the video stream or capturing target images, acquisition can be performed under different external factors such as pose angles, occlusions, and lighting conditions.
[0091] In addition, the image acquisition subunit also includes a filtering subunit, which performs object detection on the acquired target training images to remove target training images that do not contain the target. It should be noted that target training images can be understood as a collection of image data. This image data can be images targeting at least one target, corresponding to each target at different angles, with different pixel colors, etc. The amount of image data is usually quite large, reaching millions. These target training images are used to train the network to complete the model construction.
[0092] The target detection unit 42 includes: a detection subunit, which inputs the acquired target training image into the target detection model to obtain the target prediction result output by the target detection model, wherein the target detection model is trained based on the target training samples and their corresponding target detection ground values.
[0093] In an optional embodiment, the target detection unit 42 includes: a detection subunit for detecting targets in the acquired target training image; and a labeling subunit for labeling the detected targets and their corresponding positions to obtain target prediction results.
[0094] Training unit 43 includes: a center point prediction subunit, which inputs the target training image into the feature extraction layer to obtain the center point position output by the feature extraction layer; an area prediction subunit, which inputs the center point position into the target area prediction layer to obtain the predicted target area output by the target area prediction layer; and a training subunit, which constructs a loss function based on the predicted target area and the true value of the target area, and terminates training based on the convergence of the loss function.
[0095] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 51, a communication interface 52, a memory 53, and a communication bus 54, wherein the processor 51, the communication interface 52, and the memory 53 communicate with each other through the communication bus 54. The processor 51 can call logical instructions in the memory 53 to execute a target detection method, which includes: performing target detection on an acquired image of a target to be detected to obtain a target detection result; inputting the target detection result into an area prediction model to obtain the target prediction area output by the area prediction model; wherein the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground truth value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result; based on the target detection result, the actual target area is obtained; and the accuracy of the target detection result is determined based on the target prediction area and the actual target area to obtain an accuracy detection result.
[0096] Furthermore, the logical instructions in the aforementioned memory 53 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0097] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the target detection method provided by the above methods. The method includes: performing target detection on an acquired target image to be detected to obtain a target detection result; inputting the target detection result into an area prediction model to obtain a target prediction area output by the area prediction model; wherein the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result; based on the target detection result, the actual target area is obtained; and the target detection result is judged based on the target prediction area and the actual target area to obtain an accuracy detection result.
[0098] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the target detection method provided by the above methods. The method includes: performing target detection on an acquired image of a target to be detected to obtain a target detection result; inputting the target detection result into an area prediction model to obtain a target prediction area output by the area prediction model; wherein the area prediction model is trained based on the target prediction result corresponding to the target training image and the ground truth value of the target area corresponding to the target training image; the area prediction model is used to predict the target prediction area based on the position coordinates extracted from the target detection result; based on the target detection result, obtaining the actual target area; and judging whether the target detection result is accurate based on the target prediction area and the actual target area to obtain an accuracy detection result.
[0099] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A target detection method, characterized in that, include: Target detection is performed on the acquired target image to obtain the target detection result; The target detection results are input into the area prediction model to obtain the target prediction area output by the area prediction model; The area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground value corresponding to the target training image. The area prediction model is used to predict the target area based on the location coordinates extracted from the target detection results. Based on the target detection results, the actual area of the target is obtained; Based on the predicted area of the target and the actual area of the target, the accuracy of the target detection result is determined to obtain the accuracy detection result. Based on the target detection results, the actual area of the target is obtained, including: Based on the target detection results, the target's coordinates are obtained; Based on the target's coordinates, the actual area of the target is calculated. The area prediction model includes: The feature extraction layer extracts features from the input target detection results to obtain the coordinates of the center point. The target area prediction layer predicts the target area based on the coordinates of the center point. Feature extraction is performed on the input target detection results to obtain the center point position coordinates, including: Based on the target detection results, target contour features are extracted; Based on the extracted contour features, the coordinates of the center point are obtained.
2. The target detection method according to claim 1, characterized in that, The prediction based on the coordinates of the center point to obtain the target predicted area includes: Based on the coordinates of the center point, the target area is predicted using a pre-fitted regression line.
3. The target detection method according to claim 1, characterized in that, Before performing feature extraction on the input target detection results, the process also includes edge processing of the target image to be detected.
4. The target detection method according to claim 1, characterized in that, The step of determining the accuracy of the target detection result based on the predicted target area and the actual target area to obtain the accuracy detection result includes: The deviation value is obtained based on the predicted target area and the pre-obtained actual target area; The deviation value is compared with a preset range to obtain the accuracy detection result.
5. The target detection method according to claim 1, characterized in that, Training the area prediction model includes: Obtain the target training image and its corresponding ground truth area; Target detection is performed on the target training image to obtain the target prediction result; Using the target prediction results as input data for training and the target area ground truth as labels, the training model is trained to obtain an area prediction model for generating target detection results.
6. A target detection device, characterized in that, include: The target detection module performs target detection on the acquired image of the target to be detected and obtains the target detection result; The area prediction module inputs the target detection result into the area prediction model to obtain the target prediction area output by the area prediction model; wherein, the area prediction model is trained based on the target prediction result corresponding to the target training image and the target area ground truth value corresponding to the target training image; the area prediction model is used to predict the target area based on the position coordinates extracted from the target detection result. The area acquisition module obtains the actual area of the target based on the target detection results; The accuracy detection module determines whether the target detection result is accurate based on the predicted target area and the actual target area, and obtains the accuracy detection result. The area acquisition module includes: The target acquisition unit obtains the coordinate position of the target based on the target detection results; The area acquisition unit calculates the actual area of the target based on its coordinate position. The area prediction module includes: The feature extraction unit extracts features from the input target detection results to obtain the coordinates of the center point. The target area prediction unit predicts the target area based on the coordinates of the center point. The feature extraction unit includes: The feature extraction subunit extracts target contour features based on the target detection results; The coordinate acquisition sub-unit obtains the center point coordinates based on the extracted contour features.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the target detection method as described in any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the target detection method as described in any one of claims 1 to 5.
9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the target detection method as described in any one of claims 1 to 5.
Citation Information
Patent Citations
Target detection method based on deep learning
CN111382787A