Target object detection method and device, computer equipment and readable storage medium

Abstract

The invention relates to a target object detection method and device, computer equipment and a readable storage medium, and the method comprises the steps: inputting a to-be-detected image into a target detection model for preliminary detection, and obtaining a preliminary detection result of a target object; inputting the preliminary detection result into a category discrimination model for prediction to obtain a probability score of a non-target object; comparing the preliminary detection result with a pre-stored feature queue to obtain a similarity score; the feature queue is used for storing non-target features in the historical detection result; correcting the probability score based on the similarity score to obtain a target score; determining a false detection result from the preliminary detection result based on the target score; and the false detection result is eliminated from the preliminary detection result to obtain a final detection result, so that the problem of poor target object detection effect is solved, the false detection recognition accuracy is improved, and the quality of the target detection result is improved.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a target object detection method, device, computer equipment and readable storage medium. Background Art

[0002] In the field of computer vision, object detection is a crucial technology. It not only promotes the rapid development of multiple fields such as face recognition, intelligent monitoring, and human-computer interaction, but is also widely used in multiple industries with broad market prospects such as security monitoring, consumer electronics, and autonomous driving. With the continuous advancement of artificial intelligence technology, especially the rise of deep learning technology, the accuracy and efficiency of object detection technology have been significantly improved.

[0003] Currently, the target detection method based on deep convolutional neural network (CNN) is the mainstream technology in this field. CNN can automatically learn and extract high-level features in images by simulating the connection mode of human brain neurons, so as to achieve accurate recognition and positioning of target objects. This type of method usually includes steps such as feature extraction, candidate region generation, classification and regression, and finally outputs the detected target category and its specific location in the image.

[0004] However, despite the remarkable achievements of CNN-based target detection technology, it still faces the problem of false detection in practical applications, that is, the algorithm mistakenly identifies non-target objects as target categories, resulting in low target detection accuracy and still unsatisfactory performance in complex scenes and high-precision systems. Summary of the invention

[0005] In this embodiment, a target object detection method, apparatus, computer equipment and readable storage medium are provided to solve the problem of poor target detection accuracy in related technologies.

[0006] In a first aspect, a target object detection method is provided in this embodiment, the method comprising:

[0007] Input the image to be detected into the target detection model for preliminary detection to obtain the preliminary detection result of the target object;

[0008] Input the preliminary detection result into the category discrimination model for prediction to obtain a probability score of a non-target object;

[0009] Comparing the preliminary detection result with a pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; modifying the probability score based on the similarity score to obtain a target score;

[0010] Based on the target score, false detection results are determined from the preliminary detection results; the false detection results are eliminated from the preliminary detection results to obtain a final detection result.

[0011] In some embodiments, the preliminary detection result is input into a category discrimination model for prediction to obtain a probability score of a non-target object, including:

[0012] The preliminary detection result includes first position information of the target object; performing an outward expansion process on the target frame in the first position information to obtain second position information;

[0013] The second position information and the image to be detected are input into a category discrimination model to predict a probability score that the preliminary detection result is a non-target object.

[0014] In some of the embodiments, the preliminary detection result is compared with a pre-stored feature queue to obtain a similarity score, including:

[0015] The preliminary detection result includes target features of the target object;

[0016] Calculating the similarity between the target feature and each of the non-target features in the feature queue;

[0017] A similarity score is obtained based on the maximum value among the similarities.

[0018] In some embodiments, the probability score is modified based on the similarity score to obtain a target score, including:

[0019] Obtaining a weight parameter; calculating the product of the similarity score and the weight parameter;

[0020] A target score is obtained based on the product and the sum of the probability scores.

[0021] In some of the embodiments, determining the false detection result from the preliminary detection result based on the target score includes:

[0022] Obtaining a preset first threshold and a preset second threshold, wherein the first threshold is less than the second threshold;

[0023] Determining a first false detection result from the preliminary detection result based on the target score being higher than the first threshold, wherein the first false detection result is used to determine the final detection result;

[0024] Based on the target score being higher than the second threshold, a second false detection result is determined from the preliminary detection result, and the second false detection result is used to update the feature queue.

[0025] In some embodiments, the process of updating the feature queue based on the second false detection result includes:

[0026] If the feature queue is full, the feature to be replaced in the feature queue is removed, and the current feature in the second false detection result is added to the feature queue; wherein the feature to be replaced is the non-target feature with the largest similarity data with the current feature;

[0027] If the feature queue is not full, the target feature is directly added to the feature queue.

[0028] In some of the embodiments, after predicting the probability score of the preliminary detection result being a non-target object based on the category discrimination model, the method further includes:

[0029] extracting initial non-target features from the preliminary detection results based on the probability scores;

[0030] A feature queue of a fixed length is created, and the initial non-target feature is added to the feature queue.

[0031] In a second aspect, a target object detection device is provided in this embodiment, and the device includes:

[0032] A preliminary detection module is used to input the image to be detected into the target detection model for preliminary detection to obtain the preliminary detection result of the target object;

[0033] A discrimination calculation module, used to predict the probability score of the preliminary detection result being a non-target object based on a category discrimination model;

[0034] A discrimination correction module, used for comparing the preliminary detection result with a pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; based on the similarity score, the probability score is corrected to obtain a target score;

[0035] The false detection elimination module is used to determine the false detection results from the preliminary detection results based on the target score; and eliminate the false detection results from the preliminary detection results to obtain the final detection results.

[0036] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the target object detection method described in the first aspect is implemented.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the target object detection method described in the first aspect is implemented.

[0038] Compared with the related art, the target object detection method, device, computer equipment and readable storage medium provided in this embodiment obtain a preliminary detection result of the target object by inputting the image to be detected into the target detection model for preliminary detection; the preliminary detection result is input into the category discrimination model for prediction to obtain a probability score of a non-target object; the preliminary detection result is compared with a pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; the probability score is corrected based on the similarity score to obtain a target score; based on the target score, the false detection result is determined from the preliminary detection result; the false detection result is eliminated from the preliminary detection result to obtain a final detection result, which solves the problem of poor target object detection effect and improves the accuracy of false detection recognition, thereby improving the quality of target detection results.

[0039] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0041] Figure 1 The hardware structure block diagram of the terminal of the target object detection method in the embodiment of the present application;

[0042] Figure 2 Schematic diagram of the process of the target object detection method in the embodiment of the present application;

[0043] Figure 3 Schematic diagram of the processing flow of the target detection model in the embodiment of the present application;

[0044] Figure 4 This is a flow chart of a target object detection method in a preferred embodiment of the present application;

[0045] Figure 5 A schematic diagram of a process for creating a feature queue in a preferred embodiment of the present application;

[0046] Figure 6 4 is a structural block diagram of a target object detection device in an embodiment of the present application.

[0047] Figure numerals: 102, processor; 104, memory; 106, transmission device; 108, input and output device; 61, preliminary detection module; 62, judgment calculation module; 63, judgment correction module; 64, false detection elimination module. DETAILED DESCRIPTION

[0048] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0049] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0050] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. For example, running on a terminal, Figure 1 FIG. 1 is a hardware structure diagram of a terminal of the target object detection method of this embodiment. Figure 1 As shown, the terminal may include one or more ( Figure 1 Only one is shown in the figure) processor 102 and memory 104 for storing data, wherein processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1The structure shown is for illustration only and does not limit the structure of the above terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0051] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the target object detection method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0052] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by the communication provider of the terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.

[0053] In this embodiment, a target object detection method is provided. Figure 2 is a flow chart of the target object detection method of this embodiment. Figure 2 As shown, the process includes the following steps:

[0054] Step S210: input the image to be detected into the target detection model for preliminary detection to obtain a preliminary detection result of the target object.

[0055] Specifically, the target detection model is a detection model pre-trained for the target object to be detected, such as a face detection model, a vehicle detection model, etc., which is used to detect various target objects in a targeted manner. The preliminary detection results include but are not limited to the detected target frame, features, etc. In one embodiment, see Figure 3, adjust the size of the image to be detected, input the adjusted image to be detected into the target detection model, and extract the target features by the basic network in the middle part of the target detection model. The final output layer learns the category position, category features, etc. according to the feature representation and supervision information of the previous layer. After the model training is completed, the target box and the corresponding feature map of the target object can be output at the same time.

[0056] Step S220, input the preliminary detection result into the category discrimination model for prediction to obtain the probability score of the non-target object.

[0057] Specifically, since there may be false detections in the preliminary detection results, a category discrimination model is used to identify false detections. The category discrimination model includes a convolutional grid, and finally classification is performed through a fully connected layer to output the probability value of each category. The target category can be determined by pre-setting a threshold. In this embodiment, the category discrimination model is set as a two-class discrimination model to determine whether the input is a non-target object.

[0058] Step S230, comparing the preliminary detection result with the pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; based on the similarity score, the probability score is corrected to obtain a target score.

[0059] Specifically, in order to improve the accuracy of false positive identification, the prediction results of the category discrimination model are corrected through the feature queue. The target features of the target object are obtained based on the preliminary detection results, and the similarity between the target features and the historical non-target features in the feature queue is calculated. The sources of historical detection results include but are not limited to detection results identified as non-target objects by the category discrimination model, or detection results identified as non-target objects by other algorithms in the same scenario.

[0060] Step S240, based on the target score, determine the false detection results from the preliminary detection results; remove the false detection results from the preliminary detection results to obtain the final detection results.

[0061] Specifically, the false detection result can also be used to update the feature queue, thereby completing the dynamic maintenance of the feature queue.

[0062] In this embodiment, an end-to-end adaptive false positive screening system is formed by the coordinated use of the category discrimination model and the target detection model. At the same time, the feature similarity is used to assist the category discrimination model in making judgments, thereby improving the accuracy of false positive recognition and thus improving the accuracy of overall target object recognition.

[0063] In some embodiments, based on the above step S220, the preliminary detection result is input into the category discrimination model for prediction to obtain a probability score of a non-target object, including:

[0064] Step S221, the preliminary detection result includes first position information of the target object; the target frame in the first position information is expanded to obtain second position information.

[0065] Step S222: input the second position information and the image to be detected into a category discrimination model to predict the probability score of the preliminary detection result being a non-target object.

[0066] Specifically, the shape, size, position and other features of the target objects in the image vary, and the original target box may not accurately cover the target object. By expanding the target box, the target is more completely contained in the detection target box, thereby improving the recognition ability of the category discrimination model for the target.

[0067] In some of the embodiments, the preliminary detection result is compared with the pre-stored feature queue to obtain a similarity score, including:

[0068] Step S310, the preliminary detection result includes the target feature of the target object; and the similarity between the target feature and each non-target feature in the feature queue is calculated.

[0069] Step S320: obtaining a similarity score based on the maximum value of the similarities.

[0070] Specifically, the length of the feature queue is k (k is a hyperparameter and can be adjusted), that is, 0-k non-target features can be stored. The k value can be set in combination with factors such as the target object or computing resources. The feature queue achieves effective dimensionality reduction of historical data while retaining most of the useful information to facilitate subsequent calculations. In addition, by comparing with each feature in the feature queue, it can ensure that the feature most similar to the target feature is found, thereby improving the accuracy of feature matching.

[0071] In some of the embodiments, the probability score is modified based on the similarity score to obtain a target score, including:

[0072] Step S410, obtaining a weight parameter; calculating the product of the similarity score and the weight parameter.

[0073] Step S420, obtaining a target score based on the product and the sum of the probability scores.

[0074] Specifically, the similarity score is added as a correlation coefficient to the output probability value of the category discrimination model, and a preset threshold is used to determine whether the final classification score meets the requirements. 分类 The calculation formula is as follows:

[0075] S classification = αTsimilar + P classification;

[0076] Among them, α represents the weight parameter, which is an adjustable hyperparameter used to adjust the similarity score T similar The weight ratio in the classification system; P 分类 The probability score output for the category classification.

[0077] Set the target score S 分类 The size is determined by the threshold and the comparison result is output. If it is a non-target, the target is eliminated.

[0078] In some of the embodiments, determining the false detection result from the preliminary detection result based on the target score includes:

[0079] Step S510, obtaining a preset first threshold and a preset second threshold, wherein the first threshold is smaller than the second threshold.

[0080] Step S520: determining a first false detection result from the preliminary detection result based on a target score higher than a first threshold, and the first false detection result is used to determine a final detection result.

[0081] Step S530: Based on the target score higher than the second threshold, a second false detection result is determined from the preliminary detection result, and the second false detection result is used to update the feature queue.

[0082] In this implementation, the dual threshold setting ensures that the feature queue only stores high-confidence target feature information, which will assist the classification model in making judgments and improve the overall accuracy of the system.

[0083] In some embodiments, the process of updating the feature queue based on the second false detection result includes:

[0084] Step S610, if the feature queue is full, the feature to be replaced in the feature queue is removed, and the current feature in the second false detection result is added to the feature queue; wherein the feature to be replaced is a non-target feature having the largest similarity data with the current feature.

[0085] Step S620: If the feature queue is not full, the target feature is directly added to the feature queue.

[0086] Specifically, the basis for judging whether the feature queue is full is whether the current length of the feature queue is less than the preset length k. If the queue length is less than k, the feature is directly added. Otherwise, the feature with the highest similarity to the current special calculation feature in the feature queue is removed, and the current target feature is updated to the feature queue. In this way, the diversity of the feature queue can be guaranteed, and only the most similar features are replaced.

[0087] In some of the embodiments, after predicting the probability score of the preliminary detection result being a non-target object based on the category discrimination model, the method further includes:

[0088] Step S710, extracting initial non-target features from the preliminary detection results based on the probability scores.

[0089] Step S720: Create a feature queue of fixed length and add the initial non-target features to the feature queue.

[0090] Specifically, the initial feature queue can be constructed based on the prediction results of the category discrimination model. After the subsequent final detection results are generated, the final detection results are used as historical detection results to dynamically update the feature queue.

[0091] The present embodiment is described and illustrated below through preferred embodiments.

[0092] Figure 4 FIG. 1 is a flow chart of the target object detection method of the preferred embodiment. Figure 4 In this preferred embodiment, the target object detection method is described and illustrated by taking the application scenario of face detection as an example.

[0093] S1. Input the image to be detected into the target detection model for preliminary detection to obtain preliminary detection results of the face; the preliminary detection results include the target frame and target features of the face.

[0094] S2. After the target frame in the detection image is expanded, the expanded target frame image is input into the category discrimination model for prediction to obtain a probability score of the target frame image being a non-face.

[0095] S3. Obtain a pre-stored feature queue, compare the target features in the preliminary detection results with the non-target features in the feature queue one by one, and calculate the similarity to assist the category discrimination model in making false detection judgments.

[0096] See also Figure 5 , the process of building a feature queue includes:

[0097] S3.1. Apply for a feature queue.

[0098] S3.2. Set the feature queue length to k (hyperparameter, adjustable).

[0099] S3.3, Dynamic maintenance: Whenever the system determines that the target is not a face, the non-face features are saved in the queue, and the length of k is always saved.

[0100] S4. Add the maximum similarity value as the correlation coefficient to the output probability score of the category discrimination model to obtain the target score S 分类 . Set thresholds ε1 and ε2, usually ε2>ε1, when the target score S 分类 >ε1, the current detection object is considered to be a non-face, and the process continues to step S5. 分类 >ε1 and S分类 >ε2, execute step S3.3 to update the current target feature into the feature queue.

[0101] S5. Eliminate non-face targets from the preliminary detection results.

[0102] In this preferred embodiment, an end-to-end adaptive face false detection screening system is formed by the coordinated use of the classification discrimination model and the face detection model. At the same time, the dynamic queue maintains features, and assists the classification discrimination model in making judgments through feature similarity, thereby improving the overall accuracy of the system. The dual threshold setting of this preferred embodiment ensures that the feature queue only stores high-confidence target feature information, which will assist the classification model in making judgments and improve the overall accuracy of the system.

[0103] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0104] In this embodiment, a target object detection device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. The terms "module", "unit", "subunit" and the like used below can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0105] Figure 6 is a structural block diagram of the target object detection device of this embodiment, such as Figure 6 As shown, the device includes: a preliminary detection module 61, a discrimination calculation module 62, a discrimination correction module 63 and a false detection elimination module 64.

[0106] The preliminary detection module 61 is used to input the image to be detected into the target detection model for preliminary detection to obtain a preliminary detection result of the target object.

[0107] The discrimination calculation module 62 is used to predict the probability score of the preliminary detection result being a non-target object based on the category discrimination model.

[0108] The discrimination correction module 63 is used to compare the preliminary detection result with the pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in the historical detection results; based on the similarity score, the probability score is corrected to obtain the target score.

[0109] The false detection elimination module 64 is used to determine the false detection results from the preliminary detection results based on the target score; and eliminate the false detection results from the preliminary detection results to obtain the final detection results.

[0110] In some of the embodiments, the discrimination calculation module 62 is also used to prepare a preliminary detection result including first position information of a target object; perform an outward expansion process on a target box in the first position information to obtain second position information; input the second position information and the image to be detected into a category discrimination model to predict a probability score of a non-target object as a preliminary detection result.

[0111] In some of the embodiments, the discrimination correction module 63 is also used for the preliminary detection results including the target features of the target object; calculating the similarity between the target features and each non-target feature in the feature queue; and obtaining a similarity score based on the maximum value of the similarities.

[0112] In some of the embodiments, the discrimination correction module 63 is further used to obtain a weight parameter; calculate the product of the similarity score and the weight parameter; and obtain a target score based on the sum of the product and the probability score.

[0113] In some of the embodiments, the false detection elimination module 64 is also used to obtain a preset first threshold and a preset second threshold, the first threshold is less than the second threshold; based on a target score higher than the first threshold, a first false detection result is determined from the preliminary detection result, and the first false detection result is used to determine the final detection result; based on a target score higher than the second threshold, a second false detection result is determined from the preliminary detection result, and the second false detection result is used to update the feature queue.

[0114] In some of the embodiments, a feature queue maintenance module is also included, which is used to remove the feature to be replaced in the feature queue if the feature queue is full, and add the current feature in the second false detection result to the feature queue; wherein the feature to be replaced is a non-target feature with the largest similarity data with the current feature; if the feature queue is not full, the target feature is directly added to the feature queue.

[0115] In some of the embodiments, the feature queue maintenance module is further used to extract initial non-target features from the preliminary detection results based on the probability scores; create a feature queue of fixed length, and add the initial non-target features to the feature queue.

[0116] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0117] In this embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0118] Optionally, the computer device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0119] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and will not be repeated in this embodiment.

[0120] In addition, in combination with the target object detection method provided in the above embodiments, a storage medium may be provided in this embodiment to implement the target object detection method. The storage medium stores a computer program; when the computer program is executed by a processor, any target object detection method in the above embodiments is implemented.

[0121] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0122] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

[0123] The term "embodiment" in this application refers to a specific feature, structure or characteristic described in conjunction with the embodiment that can be included in at least one embodiment of the present application. The appearance of this phrase in various locations in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. It is clearly or implicitly understood by those of ordinary skill in the art that the embodiments described in this application can be combined with other embodiments without conflict.

[0124] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of patent protection. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the attached claims.

Claims

1. A target object detection method, characterized in that: The method comprises: Input the image to be detected into the target detection model for preliminary detection to obtain the preliminary detection result of the target object; Input the preliminary detection result into the category discrimination model for prediction to obtain a probability score of a non-target object; Comparing the preliminary detection result with a pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; modifying the probability score based on the similarity score to obtain a target score; Based on the target score, false detection results are determined from the preliminary detection results; the false detection results are eliminated from the preliminary detection results to obtain a final detection result.

2. The target object detection method according to claim 1, characterized in that: The preliminary detection results are input into the category discrimination model for prediction to obtain the probability score of the non-target object, including: The preliminary detection result includes first position information of the target object; performing an outward expansion process on the target frame in the first position information to obtain second position information; The second position information and the image to be detected are input into a category discrimination model to predict a probability score that the preliminary detection result is a non-target object.

3. The target object detection method according to claim 1, characterized in that: Compare the preliminary detection results with the pre-stored feature queue to obtain a similarity score, including: The preliminary detection result includes target features of the target object; Calculating the similarity between the target feature and each of the non-target features in the feature queue; A similarity score is obtained based on the maximum value among the similarities.

4. The target object detection method according to claim 3, characterized in that: Modifying the probability score based on the similarity score to obtain a target score includes: Obtaining a weight parameter; calculating the product of the similarity score and the weight parameter; A target score is obtained based on the product and the sum of the probability scores.

5. The target object detection method according to claim 1, characterized in that: Determining a false detection result from the preliminary detection result based on the target score includes: Obtaining a preset first threshold and a preset second threshold, wherein the first threshold is less than the second threshold; Determining a first false detection result from the preliminary detection result based on the target score being higher than the first threshold, wherein the first false detection result is used to determine the final detection result; Based on the target score being higher than the second threshold, a second false detection result is determined from the preliminary detection result, and the second false detection result is used to update the feature queue.

6. The target object detection method according to claim 5, characterized in that: The process of updating the feature queue based on the second false detection result includes: If the feature queue is full, the feature to be replaced in the feature queue is removed, and the current feature in the second false detection result is added to the feature queue; wherein the feature to be replaced is the non-target feature with the largest similarity data with the current feature; If the feature queue is not full, the target feature is directly added to the feature queue.

7. The target object detection method according to claim 1, characterized in that: After predicting the probability score of the preliminary detection result being a non-target object based on the category discrimination model, the method further includes: extracting initial non-target features from the preliminary detection results based on the probability scores; A feature queue of a fixed length is created, and the initial non-target feature is added to the feature queue.

8. A target object detection device, characterized in that: The device comprises: A preliminary detection module is used to input the image to be detected into the target detection model for preliminary detection to obtain the preliminary detection result of the target object; A discrimination calculation module, used to predict the probability score of the preliminary detection result being a non-target object based on a category discrimination model; A discrimination correction module, used for comparing the preliminary detection result with a pre-stored feature queue to obtain a similarity score; the feature queue is used to store non-target features in historical detection results; based on the similarity score, the probability score is corrected to obtain a target score; The false detection elimination module is used to determine the false detection results from the preliminary detection results based on the target score; and eliminate the false detection results from the preliminary detection results to obtain the final detection results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images