Target detection method, device, system and program product

Abstract

The invention relates to a target detection method, device and system and a program product, and relates to the technical field of target detection. The target detection method comprises the following steps: determining a target area according to a to-be-detected image; performing first target detection on the target area; in response to the fact that the target is not detected in the target area, dividing the target area to obtain a plurality of sub-target areas; performing second target detection on each sub-target region in the plurality of sub-target regions; and in response to the target detected in the plurality of sub-target areas, determining position information of the target.

Description

Technical Field

[0001] This disclosure relates to the field of target detection technology, and in particular to a target detection method, apparatus, system and program product. Background Technology

[0002] With the rapid development of the field of artificial intelligence, object detection is an important technical means in the application of artificial intelligence. How to accurately and efficiently detect objects is closely related to whether artificial intelligence can be implemented. Summary of the Invention

[0003] One of the technical problems this disclosure aims to solve is how to improve the efficiency and accuracy of target detection.

[0004] According to some embodiments of the first aspect of this disclosure, a target detection method is provided, comprising: determining a target region based on an image to be detected; performing a first target detection on the target region; dividing the target region into multiple sub-target regions in response to no target being detected in the target region; performing a second target detection on each of the multiple sub-target regions; and determining the location information of the target in response to a target being detected in the multiple sub-target regions.

[0005] In some embodiments, in response to detecting a target in multiple sub-target regions, determining the target's location information includes: determining the sub-target region where the target is located and a first location information of the target in the sub-target region; determining a second location information of the target in the target region based on the first location information and the location information between the sub-target region and the target region; and determining the location information based on the second location information.

[0006] In some embodiments, the positional information between the sub-target region and the target region is determined based on the relationship between the positional information of the reference position corresponding to the sub-target region in the sub-target region and the positional information of the reference position in the target region, and different sub-target regions correspond to different reference positions.

[0007] In some embodiments, dividing the target region into multiple sub-target regions includes: dividing the target region according to the size ratio between the target and the target region.

[0008] In some embodiments, there is an overlapping region between every two adjacent sub-target regions in the plurality of sub-target regions, wherein the size of the overlapping region is determined according to the size of the target.

[0009] In some embodiments, determining the target region based on the image to be detected includes: determining a feature map of the image to be detected using a target detection model; presetting multiple anchor boxes on the feature map; and determining the target region within the multiple anchor boxes.

[0010] In some embodiments, the target detection method further includes: generating control instructions based on the target's location information, wherein the control instructions are used to control the target.

[0011] In some embodiments, generating control commands based on the target's location information includes: determining the target's angle information based on the target's location information and the center location information of the target area; and generating control commands based on the angle information.

[0012] In some embodiments, the target detection method further includes: recording a first number of failures in response to the failure to successfully determine a target region; and updating the image to be detected in response to the first number of failures being less than or equal to a first threshold.

[0013] In some embodiments, the target detection method further includes: recording a second failure count in response to no target being detected in a plurality of sub-target regions; and updating the image to be detected in response to the second failure count being less than or equal to a second threshold.

[0014] In some embodiments, the target detection method further includes: issuing a fault alarm in response to a first failure count exceeding a first threshold or a second failure count exceeding a second threshold.

[0015] According to some embodiments of the second aspect of this disclosure, a target detection apparatus is provided, comprising: a first determining unit configured to determine a target region based on an image to be detected; a first detecting unit configured to perform a first target detection on the target region; a segmenting unit configured to segment the target region to obtain a plurality of sub-target regions in response to no target being detected in the target region; a second detecting unit configured to perform a second target detection on each of the plurality of sub-target regions; and a second determining unit configured to determine the location information of the target in response to a target being detected in the plurality of sub-target regions.

[0016] According to some embodiments of the third aspect of this disclosure, a target detection apparatus is provided, including: a memory and a processor coupled to the memory, the processor being configured to execute the target detection method of any of the above embodiments based on instructions stored in the memory.

[0017] According to some embodiments of the fourth aspect of this disclosure, a target detection system is provided, including a target device in any of the above embodiments; and a sensor configured to acquire and send an image to be detected to the target detection device.

[0018] According to some embodiments of the fifth aspect of this disclosure, a computer-readable storage medium is provided having computer instructions stored thereon that, when executed by a processor, implement the target detection method of any of the above embodiments.

[0019] According to some embodiments of the sixth aspect of this disclosure, a computer program product is provided, including computer instructions that, when executed by a processor, implement the target detection method of any of the above embodiments.

[0020] In the above embodiments, a hierarchical target detection method is provided. This method prioritizes global target detection, and only performs local target detection when global detection fails to detect a target within the target region. Considering that targets may be large or small, large targets can be directly detected globally, while small targets can be detected locally. Given that small targets occupy a small proportion of the target region, have inconspicuous features, and are difficult to detect, dividing the target region into multiple smaller sub-target regions effectively improves the proportion of targets within these sub-target regions, overcoming the difficulty of detection to some extent. This hierarchical target detection method improves the efficiency and accuracy of target detection and also automates it. Attached Figure Description

[0021] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0022] This disclosure can be more clearly understood with reference to the accompanying drawings and the following detailed description.

[0023] Figure 1 Schematic diagrams illustrating some embodiments of the target detection method of this disclosure are shown.

[0024] Figure 2 Schematic diagrams illustrating some embodiments of the components of this disclosure for acquiring an image to be detected.

[0025] Figure 3 Schematic diagrams illustrating some embodiments of the first target detection process of this disclosure.

[0026] Figure 4 The diagram illustrates some embodiments of the process of dividing the target region and detecting the second target as disclosed in this disclosure.

[0027] Figure 5 Schematic diagrams illustrating other embodiments of the target detection method of this disclosure are shown.

[0028] Figure 6 Schematic diagrams illustrating some embodiments of the location information of the target of this disclosure.

[0029] Figure 7 Schematic diagrams illustrating further embodiments of the target detection method of this disclosure.

[0030] Figure 8Schematic diagrams illustrating further embodiments of the target detection method of this disclosure.

[0031] Figure 9 Schematic diagrams illustrating further embodiments of the target detection method of this disclosure.

[0032] Figure 10 Schematic diagrams illustrating some embodiments of the target detection apparatus of this disclosure.

[0033] Figure 11 Schematic diagrams showing other embodiments of the target detection apparatus of this disclosure are shown.

[0034] Figure 12 Schematic diagrams illustrating some embodiments of the target detection system of this disclosure are shown. Detailed Implementation

[0035] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0036] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0037] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0038] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0039] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0040] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0041] With the rapid development of the automotive industry, the demand for automation and intelligentization in tire manufacturing is becoming increasingly urgent. In the tire assembly process, accurate positioning of the valve core is crucial for proper tire installation and inflation. Therefore, precise positioning of the valve core is a prerequisite for automation and intelligentization in tire manufacturing. Improving the efficiency and accuracy of target inspection has become a pressing issue.

[0042] Therefore, improving the efficiency and accuracy of target detection is a problem that needs to be solved. To address this issue, this disclosure proposes a target detection method, as detailed below.

[0043] Figure 1 Schematic diagrams illustrating some embodiments of the target detection method of this disclosure are shown.

[0044] like Figure 1 As shown, the target detection method includes steps 110 to 150, and the target detection method is performed by the target detection device.

[0045] In step 110, the target region is determined based on the image to be detected.

[0046] Taking the valve core as the target for testing as an example, Figure 2 The diagram illustrates some embodiments of the components for acquiring an image to be detected according to the present disclosure, wherein the valve core is the core component of the tire inflation port, and is typically a small metal nozzle-shaped structure.

[0047] like Figure 2 As shown, the components for acquiring the image to be detected include an industrial computer 21, an image sensor 22, a tire 23, a conveyor belt 24, and a supplementary light 25. The industrial computer 21 can be a target detection device, and the field of view 221 of the image sensor 22 is the area that the image sensor 22 can acquire, i.e., the tire area included in the image to be detected. The image sensor can be an industrial camera, webcam, or other similar device.

[0048] The industrial computer 21 is mainly used to receive the image to be detected sent by the image sensor 22 and perform target detection on the image. Target detection can also be understood as valve core detection. For example, the industrial computer 21 can process and analyze the image to be controlled to output the valve core position information, generate and send control commands to the control equipment to guide the control equipment to perform subsequent operations. The industrial computer 21 can obtain the image to be detected collected by the image sensor 22 by calling a specific toolkit. The valve core position information can be understood as the spatial coordinates of the valve core. The control equipment can be, for example, a robotic arm, and the subsequent operations performed by the control equipment can be, for example, tire assembly and tire inflation.

[0049] Image sensor 22 is mainly used to acquire high-definition images of tires on the conveyor belt, providing raw data for target detection, where the raw data is equivalent to the image to be detected.

[0050] Field of view 221 is primarily used to ensure that the image to be detected completely covers the rim area and its interior, guaranteeing that the rim area and its interior are always within the effective detection range. For example, the rim area and its interior of the tire can be used as the target area. Here, the rim refers to the metal ring part of the tire that contacts the hub, which usually has a regular geometric shape, making it easy to locate and identify.

[0051] Tire 23 is the object of raw data collection, and the target is the valve core on tire 23.

[0052] The conveyor belt 24 is mainly used to transport the tire to the preset station directly below the image sensor 22 to achieve the continuity of image acquisition and thus facilitate the continuity of target detection.

[0053] The supplementary light 25 is mainly used to provide a stable and uniform lighting environment for the tire, reduce interference from external light, and ensure the quality of the image to be inspected and the stability of target detection.

[0054] Considering the interference factors such as changes in lighting or stains during target detection, and especially if the target is the valve core of a tire, the tire surface texture is complex and the reflectivity is inconsistent; these factors directly affect the accuracy of target detection. During the acquisition of the image to be detected, a supplementary light is used to provide a stable and uniform lighting environment for the tire, overcoming fluctuations in ambient lighting and effectively improving the accuracy and robustness of target detection.

[0055] For example, based on the image to be detected, the target region can be determined. This can be quickly identified using a deep learning object detection model, which can also be called an object detection model. If the object being collected is a tire, and the target is the valve core, then the image to be detected is the tire image. This image needs to include the complete rim of the tire, and the target region is the rim area and the area inside the rim. Alternatively, the deep learning object detection model could be a single-stage object detection algorithm (YOLO, You Only Look Once), such as YOLOv5. YOLOv5 is a single-stage object detection algorithm that is fast and accurate.

[0056] By cropping the target region from the image to be detected, where the target object is a tire (for example, the target region can also be understood as the rim's Region of Interest (ROI), background interference can be effectively reduced in the subsequent first and second target detection processes, thereby improving the efficiency and accuracy of subsequent target detection.

[0057] In step 120, the target region is subjected to first target detection.

[0058] Taking a tire as the data collection object as an example, the target area is the rim ROI. The first target detection can be understood as detecting the valve core in the rim ROI. The first target detection can be performed by the valve core detection model.

[0059] The process of first target detection involved in step 120 is described below with reference to the following embodiments. Figure 3 Schematic diagrams illustrating some embodiments of the first target detection process of this disclosure.

[0060] like Figure 3 As shown, the first target detection process mainly includes two parts, with Figure 3 The following explanations are provided for Figures A and B. Figure 3 Figure A in the diagram represents the target area, which includes only the rim area and its internal area. Figure 3 Figure B in the diagram shows the location of the target in the target area, where position 31 is the location of the target in the target area, and the target in the target area is the valve core on the wheel rim.

[0061] In step 130, in response to the absence of a target in the target region, the target region is divided to obtain multiple sub-target regions.

[0062] If the target region is not detected, dividing the target region into multiple sub-target regions can help with subsequent detection of a second target in the sub-target regions.

[0063] If a target is detected in the target area, there is no need to divide the target area; the target's location information can be determined directly.

[0064] Multiple sub-target regions can be represented as F sub F subi This represents the i-th sub-target region, where i ranges from 1 to n, and n represents the total number of sub-target regions.

[0065] In step 140, a second target detection is performed on each of the multiple sub-target regions.

[0066] The first target detection and the second target detection can be the same or different. The same second target detection can be performed on each sub-target region, or different second target detection can be performed on each sub-target region.

[0067] For example, performing second target detection on each sub-target region can be done in parallel or sequentially.

[0068] By performing secondary target detection on each sub-target region, the success rate of target detection in complex environments is improved, and the probability of target detection is increased to a certain extent.

[0069] If no target is detected in the target region in the first target detection, the process of dividing the target region involved in step 130 and the process of second target detection involved in step 140 are described below in conjunction with the following embodiments. Figure 4 The diagram illustrates some embodiments of the process of dividing the target region and detecting the second target as disclosed in this disclosure.

[0070] like Figure 4 As shown, with Figure 4 The process of dividing the target region and the process of second target detection are described using diagrams A, B, C, and D. Figure 4 Figure A in the diagram represents the target area, which includes only the rim area and its internal area. Figure 4 Figure B in the diagram shows multiple sub-target regions, namely sub-target region Q1, sub-target region Q2, sub-target region Q3 and sub-target region Q4, which are determined by dividing the target region. Figure 4 Figure C in the diagram shows a sub-target region including the target and the target's location within that sub-target region.41 Figure 4 The D diagram in the figure shows the location of the target in the target area 42.

[0071] The target's position within a sub-target region can be its coordinates within that sub-target region, and the target's position within the target region can be its coordinates within the target region.

[0072] Figure 5 Schematic diagrams illustrating other embodiments of the target detection method of this disclosure are shown. Figure 5 It describes the logic between the first object detection and the second object detection, as well as how the second object detection specifically performs object detection on multiple sub-object regions.

[0073] like Figure 5 As shown, the target detection method includes steps 501 to 513.

[0074] In step 501, target detection begins.

[0075] In step 502, the target area is determined. Taking the valve core as an example, the target area is the rim ROI. Step 502 is equivalent to... Figure 1 Step 110 in the process.

[0076] In step 503, a first target detection is performed on the target region, wherein step 503 is equivalent to Figure 1 Step 120 in the process.

[0077] If the first target detection successfully detects the target in the target area, proceed directly to step 511; if the first target detection fails to detect the target in the target area, proceed to step 504.

[0078] In step 504, the target region is divided into multiple overlapping sub-target regions F. sub In this case, step 504 is equivalent to Figure 1 Step 130 in the process.

[0079] In some embodiments, step 140 includes Figure 5 Steps 505 to 509 in the process.

[0080] In step 505, for the i-th sub-target region F subi Perform a second target detection, where the value of i ranges from 1 to n, n is the number of sub-target regions, i represents the index of the sub-target region, and the initial value of i is 1.

[0081] In step 506, it is determined whether a target is detected in the i-th sub-target region. Taking the valve core as the target, it is determined whether the valve core is detected in the i-th sub-target region.

[0082] If a target is detected in the i-th sub-target region, proceed directly to step 510; if no target is detected in the i-th sub-target region, proceed to step 507.

[0083] In step 507, it is determined whether i is equal to n.

[0084] If i equals n, proceed to step 508; if i does not equal n, proceed to step 509 again.

[0085] In step 508, no target was detected in any of the sub-target regions, and step 513 was executed directly thereafter.

[0086] In step 509, i is incremented by one, i.e., i = i + 1, and then step 505 is executed again.

[0087] In the above embodiments, the target is detected by a hierarchical approach. First, there is the first target detection, which is the global target detection. If the target is not detected by the first target detection, considering that the target may be too small to be detected, the target area is divided and the second target detection, which is the local target detection, is performed on the sub-target areas. This overcomes the problem of the target being too small to be detected to a certain extent and improves the accuracy and efficiency of target detection.

[0088] Figure 5 The subsequent steps 510 to 513 will be described in detail in relation to step 150.

[0089] In step 150, in response to the detection of a target in multiple sub-target regions, the target's location information is determined.

[0090] The target's location information can be either its pixel coordinates within the target region or its spatial coordinates. The pixel coordinates within the target region are two-dimensional coordinates, determined through either first or second target detection. The spatial coordinates are obtained by transforming the pixel coordinates within the target region.

[0091] In some embodiments, step 150 includes steps 510 and 511.

[0092] In step 510, the coordinate information of the target in the i-th sub-target region is recorded.

[0093] In step 511, the coordinate information of the target in the target area is determined.

[0094] If the target location information determined in step 150 refers to the target's spatial coordinates, then step 150 also includes... Figure 5 Step 512 in the process.

[0095] In step 512, the coordinate information of the target in the target area is converted into the spatial coordinate information of the target, and the spatial coordinate information of the target is output. The spatial coordinate information of the target is three-dimensional coordinate information.

[0096] In step 513, the target detection ends.

[0097] Taking the valve core as the target for testing as an example, Figure 6 Schematic diagrams illustrating some embodiments of the location information of the target of this disclosure.

[0098] Figure 6 An image 60 to be detected is shown, wherein region 61 in the image 60 to be detected is the target region, position 62 in the target region 61 is the center position of the target region 61, and position 63 in the target region is the position of the target detected in the target region 61.

[0099] For example, the pixel coordinates of the center position of the target area are (1999, 1449), the pixel coordinates of the target in the target area are (2393, 2121), and the angle of the target relative to the reference line is -59.62°. The reference line passes through the center position of the target area and is in a horizontal direction to the right.

[0100] In the above embodiments, a hierarchical target detection method is provided. This method prioritizes global target detection, and only performs local target detection when global detection fails to detect a target within the target region. Considering that targets may be large or small, large targets can be directly detected globally, while small targets can be detected locally. Given that small targets occupy a small proportion of the target region, have inconspicuous features, and are difficult to detect, dividing the target region into multiple smaller sub-target regions effectively improves the proportion of targets within these sub-target regions, overcoming the difficulty of detection to some extent. This hierarchical target detection method improves the efficiency and accuracy of target detection and also automates it.

[0101] Compared to manual inspection, this method improves the efficiency of target detection and can adapt to the high-speed cycle requirements of automated production lines. Furthermore, considering that manual inspection relies heavily on human experience and is susceptible to subjective factors, resulting in a higher error rate, the aforementioned tiered target detection method effectively improves the accuracy of target detection. This overcomes, to some extent, the efficiency and accuracy issues of manual inspection, reduces reliance on manual labor, lowers labor costs, and enhances production consistency. In addition, this tiered target detection method is compatible with various tire specifications without requiring adjustments to parameters specific to different tire sizes. These parameters refer to image sensor parameters, program parameters, and the size of overlapping areas in sub-target regions. Different tire sizes can include, for example, tires with different types of valve cores.

[0102] The following examples illustrate how to determine the target area.

[0103] In some embodiments, determining the target region based on the image to be detected includes: determining a feature map of the image to be detected using a target detection model; presetting multiple anchor boxes on the feature map; and determining the target region within the multiple anchor boxes.

[0104] For example, the region corresponding to the anchor frame with the highest confidence among multiple anchor frames can be determined as the target region.

[0105] A method for determining target regions in an image to be detected is provided, making target detection feasible.

[0106] If a target is not detected in the target area, the target area needs to be divided. The following example describes how to divide the target area.

[0107] In some embodiments, dividing the target region into multiple sub-target regions includes: dividing the target region according to the size ratio between the target and the target region.

[0108] If the size ratio between the target and the target area is greater than the first ratio threshold, the number of sub-target areas after division is the first number. If the size ratio between the target and the target area is less than or equal to the first ratio threshold, the number of sub-target areas after division is the second number, where the second number is greater than the first number.

[0109] In other words, the number of sub-target regions is inversely related to their size ratio, for example, an inverse relationship, where the size ratio is the size ratio between the target and the target region.

[0110] When the size ratio between the target and the target area is small, it means that there is a large difference between the size of the target and the size of the target area. In this case, the target area is divided into a larger number of sub-target areas to enlarge the size of the target and make it easier to detect the target.

[0111] In the process of dividing the target region, the impact of the size ratio between the target and the target region on the number of sub-target regions was taken into account. Determining the number of sub-target regions based on the size ratio between the target and the target region helps to accurately determine the number of sub-target regions, reduces the risk of the number of sub-target regions being too large or too small affecting the accuracy of target detection, and improves the accuracy and efficiency of target detection.

[0112] Regarding the division of the target region, in some embodiments, there is an overlapping area between every two adjacent sub-target regions among the multiple sub-target regions, wherein the size of the overlapping area is determined according to the size of the target.

[0113] In some embodiments, the size of the overlapping region is positively correlated with the size of the target, for example, in a direct proportional relationship. The larger the target size, the larger the overlapping region; the smaller the target size, the smaller the overlapping region.

[0114] Considering that dividing the target region might result in a target simultaneously occupying two different sub-target regions, leading to target failure and directly impacting the accuracy of target detection within those sub-target regions, a new method is used. This method ensures that the target is completely contained within a single sub-target region, improving the accuracy of target detection. The overlapping area between any two adjacent sub-target regions is determined by the size of the overlapping area, which in turn depends on the target's size.

[0115] If the target is located in the overlapping area between two adjacent sub-target areas, then either of these two adjacent sub-target areas can be identified as the sub-target area where the target is located.

[0116] The following embodiments describe a target detection method for situations where the target region is not successfully determined or the target is not successfully detected in multiple sub-target regions.

[0117] In some embodiments, in response to the failure to successfully determine the target region, a first number of failures is recorded; in response to the first number of failures being less than or equal to a first threshold, the image to be detected is updated.

[0118] For example, a failure counter is established for identifying the target area, and the first failure count is recorded cumulatively through this failure counter. The first failure count is recorded by cumulative counting. For example, when the target area is identified for the first time and fails, the first failure count is recorded as 1; when the target area is identified for the second time and fails, the first failure count is recorded as 2, and so on.

[0119] Considering that the positional relationship between the acquisition object and the image sensor may cause the image sensor to fail to acquire a complete image of the acquisition object, for example, dead zones caused by highlights are difficult for the image sensor to acquire, updating the image to be detected can be understood as adjusting the position or state of the acquisition object to reduce the possibility of dead zones and obtain an updated image to be detected.

[0120] By recording the number of times the target region was not successfully identified, the target region was repeatedly identified by updating the image to be detected when the number of failures was small. This improved the accuracy of target detection and reduced the risk of incorrect target region identification due to unsuccessful identification affecting the accuracy of target detection.

[0121] In some embodiments, in response to no target being detected in multiple sub-target regions, a second failure count is recorded; in response to the second failure count being less than or equal to a second threshold, the image to be detected is updated.

[0122] For example, a failure counter can be established for detecting targets in multiple sub-target regions, and the number of second failures can be accumulated through this failure counter. The number of second failures is recorded by cumulative counting. For example, when the first target detection fails, the number of second failures is recorded as 1; when the second target detection fails, the number of second failures is recorded as 2, and so on.

[0123] By recording the number of times a target is not detected in multiple sub-target regions, and updating the image to be detected when the number of failures is small, the target is repeatedly detected in multiple target regions to improve the accuracy of target detection and reduce the risk of incorrect target identification due to the failure to detect the target, which affects the accuracy of target detection.

[0124] In some embodiments, a fault alarm is triggered in response to a first failure count exceeding a first threshold or a second failure count exceeding a second threshold.

[0125] For example, the first threshold and the second threshold can be the same or different, and the first threshold and the second threshold can be equal to 3.

[0126] When the number of times the target area is not successfully identified is large, or when the target is not detected in multiple sub-target areas a large number of times, a fault alarm should be issued in a timely manner. This allows relevant personnel to intervene and adjust the collection object and target detection process in a timely manner, which helps to improve the accuracy and efficiency of target detection, reduces the risk of getting stuck in an infinite loop of repeatedly identifying target areas or repeatedly detecting targets in multiple sub-target areas, and also ensures the safety of the target detection process.

[0127] By recording the first and second failure counts, and determining the subsequent steps—updating the image to be detected or triggering an alarm—based on the first or second failure count, a mechanism can be better adapted to the uncertainties of industrial environments.

[0128] Figure 7 Schematic diagrams illustrating further embodiments of the target detection method of this disclosure. Figure 7 This describes the situations where the target area was not successfully identified and the target was not successfully detected, as detailed below.

[0129] like Figure 7 As shown, the target detection method includes steps 701 to 716.

[0130] In step 701, target detection begins.

[0131] In some embodiments, steps 702 to 703 need to be performed before step 110 in order to acquire the image to be detected.

[0132] In step 702, the parameters are initialized, including initializing the parameters for acquiring the image to be detected, initializing the parameters for the specific tool for acquiring the image to be detected, and initializing the parameters for the target detection process. The parameters for the image to be detected may be, for example, the exposure parameters and gain parameters of the image sensor.

[0133] In step 703, the image to be detected is acquired.

[0134] In step 704, the target area is determined based on the image to be detected. Taking the valve core as the target for detection as an example, the target area is the rim area and its internal area. Step 704 is equivalent to... Figure 1 Step 110 in the process.

[0135] In some embodiments, steps 705 to 707 are performed between steps 110 and 120, mainly to describe the cases where the target area is successfully determined and the target area is not successfully determined.

[0136] In step 705, it is determined whether the target area has been successfully identified. Taking the valve core as an example, this is equivalent to determining whether the tire rim has been successfully identified.

[0137] If the target area is successfully identified, proceed to step 706; if the target area is not successfully identified, proceed directly to step 707.

[0138] In step 706, the target area is cut. Taking the valve core as an example, this is equivalent to cutting the ROI of the wheel rim. Then, step 708 is executed directly.

[0139] In step 707, the first failure count is recorded, where the first failure count is the number of times the target area was not determined. Then, step 713 is executed directly.

[0140] In step 708, target detection is performed in the target region, wherein the target detection includes first target detection and second target detection. Step 708 includes... Figure 1 Steps 120 to 140 in the process.

[0141] In some embodiments, steps 709 and 710 are performed between steps 140 and 150, primarily to describe the situation where target detection is unsuccessful.

[0142] In step 709, it is determined whether the target detection was successful, that is, whether the target was detected by the second target detection.

[0143] If the target detection is successful, proceed directly to step 711; if the target detection is unsuccessful, proceed to step 710.

[0144] In step 710, the second failure count is recorded, where the second failure count is the number of times the target detection fails. Then, step 713 is executed directly.

[0145] In step 711, the target's location information is determined, whereby the target's location information can be two-dimensional coordinates of the target within the target area. Step 711 is equivalent to... Figure 1 Step 150 in the process.

[0146] In step 712, control commands are generated and output based on the target's location information, and then step 716 is executed directly.

[0147] Combining steps 713 to 715, describe the cases where the first or second number of failures exceeds the limit and the cases where the first or second number of failures does not exceed the limit.

[0148] In step 713, it is determined whether the first number of failures or the second number of failures exceeds the limit.

[0149] If the first number of failures exceeds the limit or the second number of failures exceeds the limit, proceed to step 714. If the first number of failures does not exceed the limit and the second number of failures does not exceed the limit, proceed directly to step 715.

[0150] In step 714, exception handling is performed, such as a fault alarm, and then step 716 is executed directly.

[0151] In step 715, the acquisition object is adjusted, and step 703 is executed again to update the image to be detected. Taking the valve core as the target for detection as an example, adjusting the acquisition object can be understood as rotating the tire.

[0152] In step 716, the target detection ends.

[0153] In the above embodiments, by recording the number of target region recognition failures and the total number of target recognition failures, it is determined whether the number of target region recognition failures exceeds a limit, and whether the number of target recognition failures exceeds a limit, thus determining whether to update the image to be detected or perform anomaly handling. This processing can improve the security of the target detection process, as well as the accuracy and efficiency of target detection.

[0154] After successfully detecting the target in multiple sub-target regions, the following example describes how to determine the target's location information.

[0155] In some embodiments, in response to detecting a target in multiple sub-target regions, determining the target's location information includes: determining the sub-target region where the target is located and a first location information of the target in the sub-target region; determining a second location information of the target in the target region based on the first location information and the location information between the sub-target region and the target region; and determining the location information based on the second location information.

[0156] If the target's location information is two-dimensional location information within the target area, then the second location information is directly determined as the target's location information. If the target's location information is three-dimensional location information, then the second location information needs to be converted into the target's three-dimensional location information, and the target's three-dimensional location information is determined as the target's location information.

[0157] For example, the two-dimensional position information of the target in the target area can be the pixel coordinate information of the target in the target area, and the three-dimensional position information of the target can be the spatial coordinate information of the target.

[0158] A method for determining the location information of a target is provided. This method determines the target's coordinates within the target region by using the target's coordinates within a sub-target region and the positional information between the sub-target region and the target region. This method combines the features of target detection from multiple sub-target regions, improving the accuracy of the determined target location information and thus enhancing the accuracy of target detection.

[0159] In some embodiments, the positional information between the sub-target region and the target region is determined based on the relationship between the positional information of the reference position corresponding to the sub-target region in the sub-target region and the positional information of the reference position in the target region, and different sub-target regions correspond to different reference positions.

[0160] For example, the reference position corresponding to the sub-target area can be the origin of the coordinate system corresponding to the sub-target area, or it can be the center position of the sub-target area.

[0161] (x offset , y offset (x) represents the target's first location information within the sub-target region. sub , y sub (x) represents the location information of the reference position corresponding to the sub-target region within the sub-target region. roi , y roi ) represents the location information of the reference position corresponding to the sub-target area in the target area. The location information of the target in the target area is determined by formula (1) and formula (2), that is, (x,y).

[0162] x = x offset + x roi - x sub (1)

[0163] y = y offset + y roi - y sub (2)

[0164] The coordinate systems of both the sub-target region and the target region are based on the top left corner as the origin, with the x-axis pointing horizontally to the right and the y-axis pointing vertically downwards.

[0165] In some embodiments, target detection can also be performed by: detecting targets by template matching; detecting targets by 3D vision detection; or embedding tags into tires using radio frequency identification tags.

[0166] In some embodiments, the position information of the target in the image to be detected is determined based on the position information of the target in the target region and the positional relationship between the target region and the image to be detected. The positional relationship between the target region and the image to be detected is determined based on the relationship between the position information of the reference position corresponding to the target region in the target region and the position information of the reference position in the image to be detected.

[0167] After determining the target's location information, the control device needs to control the target through control commands. The following embodiment describes how the control device controls the target.

[0168] Figure 8 Schematic diagrams illustrating further embodiments of the target detection method of this disclosure.

[0169] like Figure 8 As shown, the target detection method includes steps 110 to 150, as well as step 810.

[0170] Steps 110 to 150 have been completed. Figure 1 A detailed description will not be provided here.

[0171] In step 810, control commands are generated based on the target's location information, wherein the control commands are used to control the target.

[0172] For example, the generated control commands can be sent to a control device, which can then control the target based on these commands.

[0173] For example, the control device could be a robotic arm.

[0174] Control commands can be generated directly based on the target's position information, or they can be generated based on the target's angle information determined by the target's position information.

[0175] Control commands are generated directly based on the target's location information. The control device can then inflate the valve core at different positions on the tire according to these commands. If the target's location information consists of pixel coordinates within the target area, these coordinates must be converted to spatial coordinates before generating the control commands. The spatial coordinate system to which the target's coordinates belong is a three-dimensional coordinate system operable by the control device, achieving an accurate mapping from two-dimensional visual information to three-dimensional operational commands and bridging two-dimensional visual information with three-dimensional physical control. If the target's location information consists of its spatial coordinates, control commands can be generated directly based on these coordinates.

[0176] By generating control commands, the detected information is directly converted into control commands, facilitating autonomous control of the target by the control equipment and enabling automation and intelligence of the control process. Furthermore, a closed-loop automated process of detection, calculation, control, and execution is achieved, realizing end-to-end automation from image acquisition to guiding the control equipment for automatic control, enhancing the practicality of target detection. In this closed-loop automated process, detection refers to target detection, calculation refers to coordinate transformation (converting two-dimensional coordinate information into three-dimensional coordinate information), control refers to the generation and transmission of control commands, and execution refers to the control equipment executing the control commands.

[0177] In some embodiments, generating control commands based on the target's location information includes: determining the target's angle information based on the target's location information and the center location information of the target area; and generating control commands based on the angle information.

[0178] When the target area corresponds to a Cartesian coordinate system, determining the target's angle information can be achieved for example: based on the target's position information and the center position information of the target area, determine the line connecting the target and the center of the target area; then, based on the angle between the connecting line and the reference line, and the quadrant of the Cartesian coordinate system to which the target belongs, determine the target's angle information. The reference line passes through the center position of the target area.

[0179] When the coordinate system corresponding to the target area is a polar coordinate system, the angle information of the target can be determined, for example, by determining the angle information of the target based on the polar angle in the target's position information, where the target's position information is the polar coordinates of the target.

[0180] Taking the reference line as a horizontal line to the right as an example, the angle information can be determined based on formulas (3) to (5).

[0181] =yy c (3)

[0182] =xx c (4)

[0183] (5)

[0184] in, This represents the angle information of the target, and the range of values ​​for the angle information is... to , (x c ,y c This indicates the center location information of the target area, which can be determined based on the width and height of the target area. This represents the difference in the vertical direction between the target's location information within the target area and the target area's center location information. This represents the difference in the horizontal direction between the target's position information within the target area and the center position information of the target area. arctan2(*) is the arctangent function for the four quadrants, ensuring quadrant consistency. Additionally, the angle information is in Cartesian coordinates. The negative sign in formula (5) is due to the opposite y-axis direction between the Cartesian coordinate system and the coordinate system containing the target area.

[0185] Considering that tires are generally assembled first and then inflated, the control commands generated based on angle information help control the equipment to ensure that the valve cores of each tire are in the same direction during tire assembly. This facilitates subsequent inflation of the valve cores in the same direction, achieving automatic tire assembly and inflation, and realizing the automation and intelligence of the tire assembly process. For example, the valve core of each tire is located directly above the center of the tire.

[0186] The target detection method is described below with reference to the following application example.

[0187] In some applications, the image sensor is mounted 2000mm above the tire, and a supplementary light can be positioned opposite the image sensor. The intrinsic parameter matrix and distortion coefficients of the image sensor are obtained using calibration tools, and these are then initialized. Hand-eye calibration can also be used to initialize the mapping between the target area's coordinate system and the spatial coordinate system. Here, hand-eye calibration is an example of "eye outside the hand," where the image sensor and the control device's base are relatively fixed. The image sensor is fixed to the frame, and the operator observes the control device's operation while the image sensor remains stationary. This establishes the relative relationship between the image sensor and the control device's base, subsequently guiding the control device's operation based on the target's spatial coordinates.

[0188] After the conveyor belt transports the tire to the preset station, the industrial control computer triggers the image sensor to acquire the tire image. The industrial control computer calls the pre-trained YOLOv5 target detection model to identify the rim ROI and output the bounding rectangle. The rim ROI of 2228×2218 pixels is cropped from the image to be detected to eliminate background interference. The acquired tire image is equivalent to the image to be detected, and the cropped rim ROI is equivalent to the target area.

[0189] For example, the YOLOv5 object detection model can be pre-trained using a training set containing 10,000 tire images with different lighting, stains, and textures. The preset workstation could be, for example, directly below the image sensor, and the coordinates of the circumscribed rectangle could be, for example, an origin at (885, 340), a width of 2228, and a height of 2218.

[0190] In some application examples, the valve core detection model is first called within the rim ROI to directly detect the rim ROI. Assuming that the valve core is successfully detected in the rim ROI, the coordinates of the valve core in the rim ROI are (2393, 2121).

[0191] Assuming the valve core is not detected within the rim ROI, the 2228×2218 rim ROI is automatically divided into four sub-target regions of size (1114 + overlap area size) × (1109 + overlap area size), where the overlap area size is 20 pixels. These sub-regions are detected in parallel. If a valve core is detected in a sub-region, the coordinates are reconstructed using the formula (x = x...). roi - x sub + x offset y = y roi -y sub + y offset This allows us to obtain the coordinate information of the valve core in the ROI of the wheel rim.

[0192] In some application examples, after determining the coordinate information of the valve core in the rim ROI, the center pixel coordinates (x, y, z) of the rim ROI are used to... c ,y c ) =(1999,1449) and the pixel coordinates (x,y)=(2393,2121) of the valve core in the rim ROI, determine that the angle between the line connecting the center position of the rim ROI and the valve core and the reference line is -59.62°. The reference line is the horizontal right direction of the x-axis, and the reference line passes through the center position of the rim ROI, as shown in formulas (6) to (8).

[0193] =yy c =672 (6)

[0194] =xx c =394 (7)

[0195] =-59.62° (8)

[0196] In some applications, the pixel coordinates (x,y)=(2393,2121) of the valve core in the rim ROI are converted into spatial coordinates in a three-dimensional coordinate system: X = 1250.3mm, Y = 820.5mm, Z = 650.2mm.

[0197] In some applications, the industrial computer sends the valve core angle information (-59.62°) and spatial coordinates (X = 1250.3 mm, Y = 820.5 mm, Z = 650.2 mm) to the control equipment. The control equipment controls the turntable to rotate the tire by -59.62°, so that the valve core faces the control equipment. In addition, the control equipment moves to the spatial coordinates of the valve core to inflate the valve core, thus completing the tire inflation operation.

[0198] In the above application example, the target detection time for each tire is less than or equal to 2 seconds, which meets the cycle time requirement of 6 tires per minute on the tire production line. Meanwhile, the accuracy rate of tire valve core detection is greater than or equal to 99.7%, and the error rate of valve core detection is less than 0.3%, which is better than the error rate of manual detection, which is 3.4%. Under complex working conditions, the detection accuracy rate is greater than or equal to 99.2%, such as the presence of stains on the tire or the tire being in strong light. The spatial coordinate error of the detected valve core is less than or equal to ±0.5mm, and the angular information error of the valve core is less than or equal to ±0.2°.

[0199] Figure 9 Schematic diagrams illustrating further embodiments of the target detection method of this disclosure. Figure 9 This describes a target detection method using valve core as an example, as detailed below.

[0200] like Figure 9 As shown, the target detection method includes steps 901 to 904.

[0201] In step 901, a complete tire image is acquired, the rim position in the tire image is determined, and the rim ROI image is cropped out. Here, the tire image is equivalent to the image to be detected, and the rim ROI is equivalent to the target area.

[0202] In step 902, the valve core is detected based on the rim ROI image to obtain the two-dimensional coordinate information of the valve core in the rim ROI, wherein the two-dimensional coordinate information can be pixel coordinate information.

[0203] In step 903, the angle information of the valve core is determined based on the two-dimensional coordinate information of the valve core in the rim ROI, and the two-dimensional coordinate information of the valve core in the rim ROI is converted into spatial coordinate information.

[0204] In step 904, the tire rotation is controlled based on the three-dimensional coordinate information and angle information of the valve core, and the robotic arm is guided to complete the tire assembly and inflation. The robotic arm is equivalent to a control device.

[0205] In the above embodiments, from acquiring tire images to controlling tire rotation, a closed-loop logical control of perception, decision-making, and execution is realized, which improves the applicability of target detection.

[0206] Figure 10 Schematic diagrams illustrating some embodiments of the target detection apparatus of this disclosure.

[0207] like Figure 10 As shown, the target detection device 1000 includes a first determination unit 1001, a first detection unit 1002, a division unit 1003, a second detection unit 1004, and a second determination unit 1005.

[0208] The first determining unit 1001 is configured to determine the target region based on the image to be detected.

[0209] The first detection unit 1002 is configured to perform first target detection on the target area.

[0210] The segmentation unit 1003 is configured to segment the target region in response to the absence of a target in the target region, thereby obtaining multiple sub-target regions.

[0211] The second detection unit 1004 is configured to perform second target detection on each of the multiple sub-target regions.

[0212] The second determining unit 1005 is configured to determine the location information of a target in response to detecting a target in multiple sub-target regions.

[0213] In the above embodiments, a target detection device is provided. This device prioritizes global target detection, and only performs local target detection when global target detection fails to detect a target within the target area. Considering that targets may be large or small, large targets can be directly detected globally, while small targets can be detected locally. Given that small targets occupy a small proportion of the target area, have inconspicuous features, and are difficult to detect, dividing the target area into multiple smaller sub-target areas effectively improves the proportion of targets within these sub-target areas, overcoming the difficulty of detection to some extent. This hierarchical target detection process improves the efficiency and accuracy of target detection and also automates it.

[0214] In some embodiments, the second determining unit 1005 is further configured to determine the sub-target region where the target is located and the first location information of the target in the sub-target region; determine the second location information of the target in the target region based on the first location information and the location information between the sub-target region and the target region; and determine the location information based on the second location information.

[0215] In some embodiments, the positional information between the sub-target region and the target region is determined based on the relationship between the positional information of the reference position corresponding to the sub-target region in the sub-target region and the positional information of the reference position in the target region, and different sub-target regions correspond to different reference positions.

[0216] In some embodiments, the partitioning unit 1003 is further configured to partition the target region according to the size ratio between the target and the target region.

[0217] In some embodiments, there is an overlapping region between every two adjacent sub-target regions in the plurality of sub-target regions, wherein the size of the overlapping region is determined according to the size of the target.

[0218] In some embodiments, the first determining unit 1001 is further configured to determine a feature map of the image to be detected by a target detection model; preset multiple anchor boxes on the feature map; and determine a target region within the multiple anchor boxes.

[0219] In some embodiments, the target detection device 1000 further includes a transmitting unit configured to generate control commands based on the target's location information, wherein the control commands are used to control the target.

[0220] In some embodiments, the sending unit is further configured to determine the angle information of the target based on the target's position information and the center position information of the target area; and to generate control commands based on the angle information.

[0221] In some embodiments, the target detection device 1000 further includes an update unit configured to record a first number of failures in response to a failed determination of a target region; and to update the image to be detected in response to a first number of failures being less than or equal to a first threshold.

[0222] In some embodiments, the update unit is further configured to record a second failure number in response to no target being detected in multiple sub-target regions; and to update the image to be detected in response to the second failure number being less than or equal to a second threshold.

[0223] In some embodiments, the target detection device 1000 further includes an alarm unit configured to issue a fault alarm in response to a first failure count exceeding a first threshold or a second failure count exceeding a second threshold.

[0224] Figure 11 Schematic diagrams showing other embodiments of the target detection apparatus of this disclosure are shown.

[0225] like Figure 11As shown, the target detection device 1000 of this embodiment includes: a memory 1101 and a processor 1102 coupled to the memory 1101. The processor 1102 is configured to execute the target detection method of any of the foregoing embodiments based on instructions stored in the memory 1101.

[0226] The memory 1101 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory stores, for example, the operating system, application programs, boot loader, and other programs.

[0227] The target detection device 1000 may also include an input / output interface 1103, a network interface 1104, and a storage interface 1105. These interfaces 1103, 1104, and 1105, as well as the memory 1101 and processor 1102, can be connected, for example, via a bus 1106. The input / output interface 1103 provides a connection interface for input / output devices such as a monitor, mouse, keyboard, touchscreen, microphone, and speakers. The network interface 1104 provides a connection interface for various networked devices. The storage interface 1105 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0228] In the above embodiments, a target detection device is provided. This device prioritizes global target detection, and only performs local target detection when global target detection fails to detect a target within the target area. Considering that targets may be large or small, large targets can be directly detected globally, while small targets can be detected locally. Given that small targets occupy a small proportion of the target area, have inconspicuous features, and are difficult to detect, dividing the target area into multiple smaller sub-target areas effectively improves the proportion of targets within these sub-target areas, overcoming the difficulty of detection to some extent. This hierarchical target detection process improves the efficiency and accuracy of target detection and also automates it.

[0229] Figure 12 Schematic diagrams illustrating some embodiments of the target detection system of this disclosure are shown.

[0230] like Figure 12 As shown, the target detection system 1200 includes the target detection device 1000 and the sensor 1201 in any of the above embodiments.

[0231] Sensor 1201 is configured to acquire and send an image to be detected to the target detection device.

[0232] In the above embodiments, a target detection system is provided that prioritizes global target detection, and only performs local target detection when global target detection fails to detect a target in the target area. Considering that targets may be large or small, large targets can be directly detected globally, while small targets can be detected locally. Given that small targets occupy a small proportion of the target area, have inconspicuous features, and are difficult to detect, dividing the target area into multiple smaller sub-target areas effectively improves the proportion of targets in the sub-target areas, overcoming the difficulty of detection to some extent. This hierarchical target detection process improves the efficiency and accuracy of target detection and also automates target detection.

[0233] In some embodiments, a computer program product is protected, comprising a computer program or instructions that, when executed by a processor, implement the target detection method described above. The computer program product includes a computer program carried on a computer-readable medium, containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network, installed from a storage device, or installed from a ROM via a target detection device. When the computer program is executed by a CPU, it performs the functions defined in the methods of the embodiments of this disclosure.

[0234] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0235] This concludes the detailed description of the target detection method, apparatus, system, and program products of this disclosure. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

[0236] The methods and systems of this disclosure may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of this disclosure are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, this disclosure may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the methods according to this disclosure. Thus, this disclosure also covers recording media storing programs for performing the methods according to this disclosure.

[0237] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A target detection method, comprising: Determine the target region based on the image to be detected; Perform a first target detection on the target region; In response to the absence of a target in the target area, the target area is divided into multiple sub-target areas; Perform second target detection on each of the plurality of sub-target regions; In response to detecting the target in the plurality of sub-target regions, the location information of the target is determined.

2. The target detection method according to claim 1, wherein, In response to detecting the target in the plurality of sub-target regions, determining the target's location information includes: Determine the sub-target region where the target is located and the first location information of the target within the sub-target region; Based on the first location information and the positional relationship between the sub-target region and the target region, the second location information of the target in the target region is determined; The location information is determined based on the second location information.

3. The target detection method according to claim 2, wherein, The positional relationship between the sub-target region and the target region is determined based on the relationship between the positional information of the reference position corresponding to the sub-target region in the sub-target region and the positional information of the reference position in the target region. Different sub-target regions correspond to different reference positions.

4. The target detection method according to claim 1, wherein, The process of dividing the target region into multiple sub-target regions includes: The target region is divided according to the size ratio between the target and the target region.

5. The target detection method according to claim 1, wherein, There is an overlapping area between every two adjacent sub-target areas in the plurality of sub-target areas, wherein the size of the overlapping area is determined according to the size of the target.

6. The target detection method according to any one of claims 1 to 5, wherein, The step of determining the target region based on the image to be detected includes: The feature map of the image to be detected is determined using a target detection model; Multiple anchor frames are preset on the feature map; The target area is determined within the plurality of anchor frames.

7. The target detection method according to any one of claims 1 to 5, further comprising: Based on the target's location information, control commands are generated, wherein the control commands are used to control the target.

8. The target detection method according to claim 7, wherein, The step of generating control commands based on the target's location information includes: Based on the target's location information and the center location information of the target area, the angle information of the target is determined; The control command is generated based on the angle information.

9. The target detection method according to any one of claims 1 to 5, further comprising: In response to the failure to successfully determine the target area, record the first failure count; In response to the first failure count being less than or equal to a first threshold, the image to be detected is updated.

10. The target detection method according to claim 9, further comprising: In response to the failure to detect the target in the plurality of sub-target regions, a second failure count is recorded; In response to the second failure count being less than or equal to the second threshold, the image to be detected is updated.

11. The target detection method according to claim 10, further comprising: A fault alarm is triggered if the first number of failures exceeds the first threshold, or if the second number of failures exceeds the second threshold.

12. A target detection device, comprising: The first determining unit is configured to determine the target region based on the image to be detected; The first detection unit is configured to perform a first target detection on the target region; A segmentation unit is configured to segment the target region to obtain multiple sub-target regions in response to the absence of a target detected in the target region. The second detection unit is configured to perform second target detection on each of the plurality of sub-target regions; The second determining unit is configured to determine the location information of the target in response to detecting the target in the plurality of sub-target regions.

13. A target detection device, comprising: Memory; and A processor coupled to the memory, the processor being configured to execute the target detection method of any one of claims 1 to 11 based on instructions stored in the memory.

14. A target detection system, comprising: The target detection device as described in claim 12 or 13; The sensor is configured to acquire and send an image to be detected to the target detection device.

15. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the target detection method according to any one of claims 1 to 11.

16. A computer program product comprising a computer program that, when executed by a processor, implements the target detection method according to any one of claims 1 to 11.

Images