Work clothes dressing detection method and equipment

Abstract

The invention discloses a work clothes dressing detection method and equipment. The method comprises the following steps: acquiring a dressing image of a to-be-detected person and at least one clothing sample image corresponding to clothing to be worn of the to-be-detected person; performing target detection on the dressing image, detecting a clothing picture part, and segmenting the clothing picture part from the dressing image; generating a clothes three-dimensional model corresponding to the clothes picture part and a work clothes three-dimensional model corresponding to each clothes sample image by using the same three-dimensional model generation network; performing similarity comparison on the clothes three-dimensional model and each work clothes three-dimensional model to obtain a comparison result; and based on a comparison result, determining whether the to-be-detected person is correctly dressed. According to the invention, the scene universality of work clothes dressing detection can be improved.

Description

Technical Field

[0001] This application relates to the field of image recognition, and in particular, to a work uniform wearing detection method and device. Background Art

[0002] To ensure operation safety, it is necessary to be able to identify and distinguish between staff and ordinary personnel. Generally, staff wear work uniform types corresponding to their job responsibilities. In related technologies, deep learning neural networks trained can be used for work uniform type recognition.

[0003] However, the deep learning neural network needs a model separately trained for different applicable scenarios to provide a relatively high-accuracy work uniform type recognition, that is, the scene generality is limited.

[0004] Application Content

[0005] The main purpose of this application is to provide a work uniform wearing detection method and device, aiming to solve the technical problem of poor scene generality of the deep learning neural network.

[0006] To achieve the above purpose, this application provides a work uniform wearing detection method, which includes:

[0007] Obtain the wearing image of the person to be detected and at least one clothing sample image corresponding to the clothing that the person to be detected should wear;

[0008] Perform object detection on the wearing image, detect the clothing picture part, and segment the clothing picture part from the wearing image;

[0009] Use the same 3D model generation network to generate the clothing 3D model corresponding to the clothing picture part and the work uniform 3D models corresponding to the respective clothing sample images;

[0010] Compare the similarity between the clothing 3D model and each of the work uniform 3D models to obtain a comparison result;

[0011] Based on the comparison result, determine whether the person to be detected is wearing the correct clothing.

[0012] In a possible embodiment of this application, the step of using the same 3D model generation network to generate the work uniform 3D models corresponding to the respective clothing sample images includes:

[0013] Perform picture mirror flipping on the clothing sample image to obtain a flipped image; the clothing sample image is a YUV format image;

[0014] Perform an interchange process on the blue chrominance component and the red chrominance component of the flipped image to obtain a flipped sample image;

[0015] Input the clothing sample image and the flipped sample image into the 3D model generation network to generate a positive-angle work uniform 3D model and a negative-angle work uniform 3D model of the same work uniform;

[0016] Both the positive-angle work uniform 3D model and the negative-angle work uniform 3D model are used as the work uniform 3D model.

[0017] In a possible embodiment of the present application, the mirror flipping of the clothing sample image to obtain a flipped image includes:

[0018] Identify a first blank span region from the clothing sample image;

[0019] Perform mirror flipping on the clothing sample image to obtain an intermediate image; the first blank span region is flipped into a second blank span region;

[0020] Based on the original orientation information of the first blank span region in the clothing sample image, perform moving and splicing processing on the second blank span region to obtain the flipped image; wherein, the orientation information of the second blank span region in the flipped image is the same as the original orientation information.

[0021] In a possible embodiment of the present application, the generation of the work uniform 3D model corresponding to each clothing sample image by using the same 3D model generation network includes:

[0022] If the clothing sample image is an original resolution image, perform clothing region detection on the clothing sample image;

[0023] If the detection result is that at least one clothing region is detected, input the target detection result and the original resolution image into the 3D model generation network to generate the work uniform 3D model corresponding to each clothing region respectively.

[0024] In a possible embodiment of the present application, the generation of the work uniform 3D model corresponding to each clothing sample image by using the same 3D model generation network includes:

[0025] If the clothing sample image is a target segmentation part segmented from the original resolution image, use the 3D model generation network to generate the work uniform 3D model of the clothing sample image.

[0026] In a possible embodiment of the present application, the comparison of the similarity between the clothing 3D model and each work uniform 3D model to obtain a comparison result includes:

[0027] Output each of the three-dimensional models of the work clothes and the model information of each of the three-dimensional models of the work clothes; the model information includes at least one of model scoring information, clothing type information, positive and negative sample information, and positive and negative angle information;

[0028] In response to a user's selection operation, determine a target three-dimensional model of work clothes from all the three-dimensional models of work clothes;

[0029] Compare the three-dimensional model of the clothing with each of the target three-dimensional models of work clothes respectively to obtain a comparison result.

[0030] In a possible embodiment of the present application, after determining whether the person to be detected is dressed correctly based on the comparison result, the method further includes:

[0031] If it is determined that the person to be detected is not dressed correctly, use the dressed image as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

[0032] In a possible embodiment of the present application, the step of using the dressed image as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear if it is determined that the person to be detected is not dressed correctly includes:

[0033] If the sample type of the clothing sample image with the highest similarity in the comparison result is a negative sample, determine that the person to be detected is not dressed correctly;

[0034] Input the dressed image into a work clothes confidence calculation model to calculate the confidence of dressing as work clothes, and obtain a confidence prediction result output by the work clothes confidence calculation model;

[0035] If the confidence prediction result is greater than a preset threshold, use the dressed image as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

[0036] In a possible embodiment of the present application, after determining whether the person to be detected is dressed correctly based on the comparison result, the method further includes:

[0037] If the dressed image meets the alarm condition, superimpose a target box reminder logo on the clothing picture part to obtain an alarm image;

[0038] Output the alarm image.

[0039] In a second aspect, the present application also provides a work clothes dressing detection device, including: a processor, a memory, and a computer program stored in the memory, and the computer program implements the steps of the work clothes dressing detection method in the first aspect when being run by the processor.

[0040] A work uniform dressing detection method proposed in an embodiment of the present application, the method includes: obtaining a dressing image of a person to be detected and at least one clothing sample image corresponding to the clothing that the person to be detected should wear; performing object detection on the dressing image to detect the clothing picture part, and segmenting the clothing picture part from the dressing image; using the same three-dimensional model generation network to generate a clothing three-dimensional model corresponding to the clothing picture part and work uniform three-dimensional models corresponding to each clothing sample image; comparing the similarity between the clothing three-dimensional model and each work uniform three-dimensional model respectively to obtain a comparison result; based on the comparison result, determining whether the person to be detected is dressed correctly.

[0041] Thus, it is not difficult to see that when performing work uniform dressing detection in an embodiment of the present application, the same set of three-dimensional model generation network is used to perform three-dimensional model modeling on the dressing image of the person to be detected and the clothing sample image corresponding to the clothing that should be worn in different applicable scenarios respectively, and then the similarity between the obtained clothing three-dimensional model and the work uniform three-dimensional model is compared to obtain the dressing detection result. In this way, since the three-dimensional model generation network does not need to be retrained for different scenarios, the scene versatility can be improved, the detection efficiency can be increased, and the detection cost can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic structural diagram of the work uniform dressing detection device of the present application;

[0043] Figure 2 It is a schematic flowchart of the first embodiment of the work uniform dressing detection of the present application;

[0044] Figure 3 It is a schematic flowchart of the second embodiment of the work uniform dressing detection of the present application;

[0045] Figure 4 It is a schematic flowchart of the third embodiment of the work uniform dressing detection of the present application.

[0046] The realization, functional features and advantages of the object of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] The dressing specifications of personnel are an important means for standardized production and life. Generally, it is required that staff and others wear work uniforms corresponding to their job responsibilities, including but not limited to work uniforms, work caps and work shoes.

[0049] Since it is time-consuming and laborious to identify the clothing on-site through the human eye or to identify specific clothing-wearing people by watching the production-site video in front of a display device, an automatic clothing-type recognition solution has emerged. For example, in a clothing-type recognition solution, after identifying a human body and obtaining an image of the human body, the recognition of work clothes is achieved by using the distance between pixel points and the color of work clothes. However, this solution can only identify whether a person is wearing work clothes, but cannot identify the specific type of work clothes.

[0050] There is also a clothing-type recognition solution based on deep learning technology, that is, using clothing sample images in the to-be-used scenario to train a clothing-type recognition network. However, the deep learning neural network needs to be trained separately for different applicable scenarios to obtain a model that can provide a relatively high accuracy in clothing-type recognition, that is, the scene generality is limited.

[0051] Therefore, this application provides a solution, that is, instead of using a clothing-type recognition network, a three-dimensional model generation network for generating a three-dimensional model based on a two-dimensional image is used. Since the same three-dimensional model generation network is applicable to work clothes in different scenarios without the need for retraining, the scene generality can be improved, and the detection efficiency can be increased and the detection cost can be reduced.

[0052] The inventive concept of this application will be further elaborated below in conjunction with some specific embodiments.

[0053] Some technologies involved in the embodiments of this application will be described below:

[0054] Computer Vision Technology (CV): Computer vision is a science that studies how to enable machines to "see". More specifically, it refers to using cameras and computers to replace the human eye to identify and measure targets, etc., which is machine vision, and further performing graphic processing to make the computer process the image into a form that is more suitable for human eyes to observe or for instruments to detect. As a scientific discipline, computer vision studies related theories and technologies, and attempts to establish an artificial intelligence system that can obtain information from images or multi-dimensional data. Computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, etc. technologies, and also includes common biometric recognition technologies such as face recognition and fingerprint recognition.

[0055] Among them, the task of object detection is to find all the objects of interest in an image, determine their categories and locations, which is one of the core problems in the field of computer vision. Due to the different appearances, shapes and poses of various objects, and the interference of factors such as illumination and occlusion during imaging, object detection has always been one of the most challenging problems in the field of computer vision.

[0056] Referring to Figure 1 , Figure 1 FIG. is a schematic structural diagram of a work clothes wearing detection device for the hardware operating environment involved in the embodiment of the present application.

[0057] As Figure 1 shown, the work clothes wearing detection device may include: a processor 1001, such as a CPU, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. Optionally, the user interface 1003 may be a display screen (Display), an input unit such as a keyboard (Keyboard), etc. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0058] It can be understood that the work clothes wearing detection device may further include a network interface 1004, and the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). Optionally, the work clothes wearing detection device may further include an RF (Radio Frequency) circuit, sensors, an audio circuit, a WiFi module, etc.

[0059] Those skilled in the art can understand that Figure 1 the structure of the work clothes wearing detection device shown in

[0060] does not limit the work clothes wearing detection device, and may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements. Figure 2 , Figure 2 FIG. shows a schematic flowchart of the first embodiment of the work clothes wearing detection method of the present application.

[0061] It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0062] In this embodiment, the work uniform wearing detection method includes:

[0063] Step S100: Obtain the wearing image of the person to be detected and at least one clothing sample image corresponding to the clothing that the person to be detected should wear.

[0064] In this embodiment, the execution entity of the work uniform wearing detection method is a work uniform wearing detection device. This work uniform wearing detection device can be connected to an image acquisition device or a camera device installed at the production site, so as to obtain the wearing image of the person to be detected collected by the image acquisition device. Among them, the person to be detected is a staff member at the production site. It can be understood that the staff member has corresponding job responsibilities on site. Thus, he or she should wear work uniforms that match his or her job responsibilities, that is, the clothing that should be worn. In addition, the production site in this embodiment can be a workshop (such as a clean workshop), a construction site, or an electric power installation site, etc., and this embodiment does not limit this.

[0065] The wearing image can be one frame or multiple frames of images collected from an image acquisition device or a camera device. Among them, the multiple frames of images can be a sequence of video frames sorted by time, or can also be disordered multiple video frames.

[0066] The clothing sample image is an image corresponding to the clothing that the person to be detected should wear. Of course, it can be understood that the clothing sample image can be a positive sample image, such as a positive sample image of a staff member correctly wearing work clothes in a work scenario, or can also be a negative sample image of a staff member not correctly wearing work clothes in a work scenario.

[0067] For the clothing sample image, before the work uniform wearing detection device starts the detection task, at least one clothing sample image can be imported into the work uniform wearing detection device by the manager of the current applicable scenario through an external device or an external interface. Or, at least one clothing sample image, that is, the work uniform sample set can also be stored in the cloud, so that the work uniform wearing detection device can obtain this work uniform sample set through the network.

[0068] Step S200: Perform object detection on the wearing image, detect the clothing picture part, and segment the clothing picture part from the wearing image.

[0069] Since the wearing image is an image frame collected at the production site, the wearing image not only includes the picture of the person to be detected, but may also include the background environment pictures of other persons to be detected or production equipment, etc. Therefore, object detection can be performed on the wearing image to extract the pixel points belonging to the clothing, so as to identify the picture areas where the corresponding human body, clothing, equipment, background, etc. are located on the wearing image. Then, according to the object detection result, image segmentation is performed on the wearing image, so as to segment out all the pixel points belonging to the clothing of the person to be detected, and obtain the clothing picture part.

[0070] It is understandable that object detection can be performed using an object detection network, and the object detection network can be pre-trained to extract clothing features on the human body. Thus, in different scenarios (such as workshops, construction sites, or outdoor scenes), the object detection network can accurately extract clothing features on the human body.

[0071] Step S300: Use the same 3D model generation network to generate the clothing 3D model corresponding to the clothing image part and the work uniform 3D models corresponding to each clothing sample image.

[0072] The 3D model generation network is a pre-trained neural network model that can perform 3D model reconstruction based on clothing images. It is understandable that due to the development of 3D model technology based on images in machine vision technology, the 3D model generation network can generate a corresponding 3D model based on a single-frame clothing image, can also generate a corresponding 3D model based on multiple-frame clothing images, and can also perform 3D modeling on some regions in the clothing image according to the input information.

[0073] Specifically, the work uniform wearing detection device can input the clothing image part into the 3D model generation network to obtain the clothing 3D model output by the 3D model generation network. Then, input the clothing sample images into the 3D model generation network to obtain the work uniform 3D models output by the 3D model generation network, so as to obtain a comparison database.

[0074] In addition, in this embodiment, before the detection task starts, the work uniform wearing detection device can load the 3D model generation network and the object detection network. After the detection task starts, the work uniform wearing detection device creates an image modeling handle, calls the 3D model generation network through the image modeling handle to generate the work uniform 3D model, and the work uniform wearing detection device creates a comparison handle, calls the object detection network and the 3D model generation network through the comparison handle to execute step S200 and step S300.

[0075] Of course, it can also be understood that the work uniform 3D models in the comparison database can be modeled in real time to dynamically add or delete the corresponding work uniform 3D models during the detection task. Of course, for the sake of improving efficiency, they can also be modeled in advance.

[0076] Step S400: Compare the similarity between the clothing 3D model and each work uniform 3D model to obtain a comparison result.

[0077] Step S500: Based on the comparison result, determine whether the person to be detected is wearing the work uniform correctly.

[0078] Perform a similarity comparison one by one between the 3D model of the clothing of the person to be detected and the 3D work clothing models of each clothing sample image in the comparison database to obtain the similarity comparison results between the 3D clothing model and each 3D work clothing model. It can be understood that the similarity comparison can also be processed using a similarity comparison neural network. This similarity comparison neural network can evaluate the similarity between two 3D models.

[0079] In a specific embodiment, when the work clothing wearing detection device executes step S400, it can output each 3D work clothing model and the model information of each 3D work clothing model; the model information includes at least one of model scoring information, clothing type information, positive and negative sample information, and positive and negative angle information; in response to the user's selection operation, determine the target 3D work clothing model from all the 3D work clothing models; perform a similarity comparison between the 3D clothing model and each target 3D work clothing model respectively to obtain the comparison results.

[0080] That is to say, when performing the 3D model comparison, the user can select the target 3D work clothing model for comparison. When selecting, the work clothing wearing detection device can display each 3D work clothing model and the model information of each 3D work clothing model through a display; the model information includes at least one of model scoring information, clothing type information, positive and negative sample information, and positive and negative angle information. Among them, the model scoring information is used to represent the model construction quality of the 3D work clothing model, such as the model score being at a high level. The clothing type information is used to represent information such as the scene usage information and clothing category of the clothing corresponding to the clothing sample image, such as a clean room work clothing. The positive and negative sample information is used to represent whether the clothing corresponding to the clothing sample image is correctly worn. If it is correctly worn, it is a positive sample; if it is not correctly worn, it is a negative sample. The positive and negative angle information is used to represent the modeling angle when performing 3D modeling on the clothing sample image. If the 3D modeling is performed at the original angle, it is a positive angle; if the modeling is performed after the image is flipped, it is a negative angle.

[0081] In an example, the display shows multiple 3D work clothing models through a window, and the background part of each 3D work clothing model also shows the model information of the 3D work model through a floating box.

[0082] Then the user can input selection information through components such as a mouse or keyboard, so as to determine the target 3D work clothing model from all the 3D work clothing models. Then the work clothing wearing detection device only performs a similarity comparison between the 3D clothing model and each target 3D work clothing model respectively to obtain the comparison results.

[0083] Of course, the user can also not make a selection, so as to perform a similarity comparison with all the 3D work clothing models.

[0084] It is not difficult to see that by allowing users to compare and filter the three-dimensional models of work clothes in the database, the process processing efficiency can be improved. In addition, through user selection, a three-dimensional model of work clothes that is more suitable for the current dressing scene of the person to be detected can be selected, thereby improving the comparison accuracy and reducing the occurrence of false alarms.

[0085] In addition, the comparison results can be displayed through a similarity comparison list, and the similarities in the similarity comparison list are sorted from large to small. In this way, the first ranked work clothes 3D model is the most similar work clothes 3D model. At this time, it can be determined whether the person to be tested is dressed correctly based on multiple top ranked work clothes 3D models, such as the first 10.

[0086] Specifically, when the clothing sample image is a positive sample image of a worker wearing work clothes correctly, if the similarity comparison with the top three-dimensional models of work clothes is greater than the preset comparison threshold, the final comparison result is a successful comparison, and it is determined that the person to be tested is correctly dressed. If the similarity comparison with the top three-dimensional models of work clothes is less than the preset threshold, it can be considered that the comparison failed and the person to be tested is not dressed correctly.

[0087] When the clothing sample image is a negative sample image of a worker not wearing work clothes correctly, if the similarity comparison with the top three-dimensional models of work clothes is greater than the preset comparison threshold, the final comparison result is successful, and it is determined that the person to be tested is not dressed correctly. If the similarity comparison with the top three-dimensional models of work clothes is less than the preset threshold, it can be considered that the comparison failed and the person to be tested is dressed correctly.

[0088] In this way, it is not difficult to see that what is used in this embodiment is not a clothing type recognition network, but a three-dimensional model generation network that generates a three-dimensional model based on a two-dimensional image, or even a model similarity comparison neural network. Since the object of this set of network models is clothing or a three-dimensional model, it is applicable to work clothes in different scenarios without the need for retraining, which can improve the scene versatility, as well as improve detection efficiency and reduce detection costs.

[0089] Of course, it can be understood that the work clothes type recognition in this embodiment can only recognize the type of work clothes corresponding to the clothing sample image, that is, it can only recognize the specific type of work clothes that the person to be detected should wear.

[0090] Based on the above embodiments, see Figure 3 , a second embodiment of the work clothes wearing detection method of the present application is proposed.

[0091] In this embodiment, step S300 specifically includes:

[0092] Step S310: Perform a screen mirror flip on the clothing sample image to obtain a flipped image.

[0093] The clothing sample image is a YUV format image.

[0094] Step 320: Swap the blue chrominance component and the red chrominance component of the flipped image to obtain a flipped sample image.

[0095] Step S330: Input the clothing sample image and the flipped sample image into a 3D model generation network to generate a positive-angle work uniform 3D model and a negative-angle work uniform 3D model of the same work uniform.

[0096] Step S340: Use both the positive-angle work uniform 3D model and the negative work uniform 3D model as the work uniform 3D model.

[0097] In this embodiment, an image mirror flip function is provided. In this way, a mirror flip operation can be performed on each frame of the clothing sample image, so as to obtain a flipped sample image corresponding to each frame of the clothing sample image.

[0098] Specifically, the clothing sample image can be an image in YUV format. Perform a screen mirror flip operation on it to obtain a flipped image. In the YUV format image, Y is the luminance component, U is the blue chrominance component, and V is the red chrominance component. It is worth mentioning that after the screen flip, attention should be paid to the arrangement order of UV in the YUV format. After the mirror flip, it becomes VU. In this embodiment, after the mirror flip, the interchange of UV is also required, that is, the blue chrominance component and the red chrominance component of the flipped image are interchanged, so as to change to the UV arrangement method.

[0099] In addition, in a specific implementation manner, when the work uniform wearing detection device executes step S310, it can also identify a first blank span area from the clothing sample image; perform a mirror flip on the clothing sample image to obtain an intermediate image, and the first blank span area is flipped into a second blank span area; based on the original orientation information of the first blank span area in the clothing sample image, perform a moving and splicing process on the second blank span area to obtain a flipped image; wherein, the orientation information of the second blank span area in the flipped image is the same as the original orientation information.

[0100] Specifically, when performing mirror flipping, it should be noted that the clothing sample image has a first blank span area. After mirror flipping, the orientation between the first blank span area and the picture area is swapped to obtain a second blank span area. For example, the first blank span area on the left side of the picture area is flipped to the second blank span area on the right side of the picture area. Therefore, it is also necessary to move and splice the part belonging to the second blank span area obtained after flipping to its proper position to keep the position of the blank span part unchanged. For example, if the right part of the clothing sample image has a blank span area, after mirror flipping, the blank span part on the left needs to be moved to the right. Thus, in this embodiment, except for the different YUV addresses, everything else is the same as before after mirror flipping.

[0101] Then, input the clothing sample image into the 3D model generation network to generate a positive-angle work uniform 3D model. And input the flipped sample image into the 3D model generation network to generate a negative-angle work uniform 3D model of the same work uniform. Among them, in order to facilitate distinguishing whether the model is a positive-angle work uniform 3D model or a negative-angle work uniform 3D model, field descriptions of the model type can be added to the obtained 3D models for easy judgment.

[0102] It is not difficult to see that in this embodiment, the same clothing sample image has model data of two different angles, thereby increasing the number of models for comparison and further improving the accuracy of comparison.

[0103] The 3D models obtained by mirror flipping modeling and the 3D models obtained by non-flipping modeling can be represented by different flag bits. When comparing, users can select all 3D models and can also freely choose the models to be used to improve the scene applicability.

[0104] As a specific implementation manner, when the work uniform wearing detection device executes step S300, if the clothing sample image is an original resolution image, then perform clothing area detection on the clothing sample image; if the detection result is that at least one clothing area is detected, then input the target detection result and the original resolution image into the 3D model generation network to generate work uniform 3D models corresponding to each clothing area respectively.

[0105] Among them, if at least one clothing area is detected in the original resolution image, the clothing sample image may include images of multiple targets. In order to only model the clothing subsequently, it is also necessary to detect all the targets that meet the preset clothing feature conditions in the picture, and then send the target information of each clothing area and the clothing sample image to the 3D model generation network. At this time, the 3D model generation network models each clothing area respectively according to the detected target information and outputs the corresponding 3D models.

[0106] Of course, the clothing sample image can also be segmented according to the target information to obtain sub-images where each clothing area is located, and then multiple sub-images are sequentially input into the 3D model generation network for modeling, and the corresponding 3D models are sequentially output to improve the modeling accuracy.

[0107] Alternatively, when the work clothing wearing detection device executes step S300, if the clothing sample image is the target segmentation part segmented from the original image, the 3D model generation network is used to generate the 3D model of the work clothing for the clothing sample image.

[0108] That is to say, when the clothing sample image is the target segmentation part segmented from the original image, that is, all or most of the pixel points in the target segmentation part belong to the same work clothing target, there is no need to perform target detection, and 3D modeling can be directly performed on the full image, thereby improving the modeling efficiency.

[0109] Based on the above embodiments, referring to Figure 4 , the third embodiment of the work clothing wearing detection method of the present application is proposed. In this embodiment, after step S500, the method further includes:

[0110] Step S600: If it is determined that the person to be detected is not dressed correctly, the dressed image is used as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

[0111] At this time, the work clothing wearing detection device can directly import the image of incorrect dressing into the database where the sample image is located. In this way, the purpose of modeling according to the actual scenario can be achieved without replacing the model, so as to improve the accuracy of subsequent detection during the continuous execution of the task.

[0112] As a specific implementation manner, when the work clothing wearing detection device executes step S600, it can determine that the person to be detected is not dressed correctly if the sample type of the clothing sample image with the highest similarity in the comparison result is a negative sample; input the dressed image into the work clothing confidence calculation model to calculate the confidence of dressing as work clothing, and obtain the confidence prediction result output by the work clothing confidence calculation model; if the confidence prediction result is greater than the preset threshold, the dressed image is used as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

[0113] Specifically, if the 3D model of the work uniform corresponding to the highest similarity ranking first in the comparison result is a negative sample image, then the detected dressed image is also a negative sample image of incorrect dressing at this time. Therefore, the dressed image can be input into the work uniform confidence calculation model to calculate the probability that it is a work uniform image, that is, confidence prediction. If the confidence prediction result is greater than the preset threshold, it is considered that the work uniform is indeed worn in the dressed image, but not correctly worn. Therefore, it needs to be used as a negative sample to update the work uniform sample set to expand the work uniform sample set. On the contrary, when the confidence prediction result is less than or equal to the preset threshold, it is considered that the dressed image does not include a work uniform, and it can be fed back to the user for manual review to determine whether the person to be detected is not wearing a work uniform, or there is no clothing area in the image, such as accidental situations like incorrect detection by the target detection network.

[0114] In addition, if the number of component images in the work uniform sample set reaches the upper limit of the database, the images can be sorted in descending order according to the confidence prediction result, and then one or more clothing sample images at the end can be deleted.

[0115] It is not difficult to see that in this embodiment, confidence calculation is required when expanding the samples to ensure that the corresponding negative samples are correctly expanded in the work uniform sample set, and to prevent the sample set from being contaminated and reducing the accuracy of subsequent detection.

[0116] Step S700: If the dressed image meets the alarm condition, a target box reminder mark is superimposed and generated on the clothing picture part to obtain an alarm image.

[0117] Step S800: Output the alarm image.

[0118] Specifically, the alarm condition can be that the detection result of the dressed image is incorrect dressing. At this time, alarm result data in a format such as json can be returned, that is, the alarm image is output. And the target box reminder information will be superimposed and displayed on the clothing picture part in the alarm image, so as to remind which person to be detected is not dressed correctly. Or, by superimposing and displaying the target box reminder mark on the clothing picture part to highlight the incorrect dressing behavior, thus playing a warning role.

[0119] For example, in an example, if the 3D model corresponding to the highest similarity ranking first in the comparison result is the 3D model corresponding to the negative sample image, it can be regarded as meeting the alarm condition.

[0120] In addition, an embodiment of the present application also provides a computer storage medium, on which a work uniform wearing detection program is stored. When the work uniform wearing detection program is executed by a processor, the steps of the work uniform wearing detection method described above are implemented. Therefore, details will not be elaborated here. In addition, the beneficial effects of the same method will not be described in detail either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application. By way of example, the program instructions can be deployed to be executed on a computing device, or on multiple computing devices located at one place, or on multiple computing devices distributed at multiple places and interconnected through a communication network.

[0121] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The above program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the above storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), or the like.

[0122] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the device embodiments provided in the present application, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement without creative efforts.

[0123] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware. Of course, it can also be implemented by dedicated hardware including application specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions accomplished by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures for implementing the same function can also be diverse, such as analog circuits, digital circuits or dedicated circuits. However, for the present application, in more cases, software program implementation is a better embodiment. Based on such an understanding, the technical solution of the present application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a floppy disk, a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disc of a computer, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present application.

[0124] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application accordingly. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present application.

Claims

1. A method for detecting the wearing of work clothes, characterized in that, The method includes: Obtaining a dressing image of the person to be detected and at least one clothing sample image corresponding to the clothing that the person to be detected should wear; Performing object detection on the dressing image to detect the clothing picture part, and segmenting the clothing picture part from the dressing image; Generating a clothing three-dimensional model corresponding to the clothing picture part and work clothing three-dimensional models corresponding to each of the clothing sample images by using the same three-dimensional model generation network; Comparing the similarity between the clothing three-dimensional model and each of the work clothing three-dimensional models to obtain a comparison result; Determining whether the person to be detected is dressed correctly based on the comparison result.

2. The work clothing dressing detection method according to claim 1, wherein The generating of the work clothing three-dimensional models corresponding to each of the clothing sample images by using the same three-dimensional model generation network includes: Performing a mirror flip on the clothing sample image to obtain a flipped image; the clothing sample image is a YUV format image; Performing an interchange process on the blue chrominance component and the red chrominance component of the flipped image to obtain a flipped sample image; Inputting the clothing sample image and the flipped sample image into the three-dimensional model generation network to generate a positive-angle work clothing three-dimensional model and a negative-angle work clothing three-dimensional model of the same work clothing; Regarding both the positive-angle work clothing three-dimensional model and the negative-angle work clothing three-dimensional model as the work clothing three-dimensional models.

3. The work clothing dressing detection method according to claim 2, characterized in that The performing of the mirror flip on the clothing sample image to obtain a flipped image includes: Identifying a first blank span area from the clothing sample image; Performing a mirror flip on the clothing sample image to obtain an intermediate image; the first blank span area is flipped into a second blank span area; Based on the original orientation information of the first blank span area in the clothing sample image, performing a moving and splicing process on the second blank span area to obtain the flipped image; wherein, the orientation information of the second blank span area in the flipped image is the same as the original orientation information.

4. The work uniform dressing detection method according to claim 1, characterized in that, The generating of the work clothing three-dimensional models corresponding to each of the clothing sample images by using the same three-dimensional model generation network includes: If the clothing sample image is an original resolution image, performing clothing area detection on the clothing sample image; If the detection result is that at least one clothing area is detected, inputting the target detection result and the original resolution image into the three-dimensional model generation network to respectively generate the work clothing three-dimensional models corresponding to each of the clothing areas.

5. The work clothing dressing detection method according to claim 1, characterized in that, The generating of the work clothing three-dimensional models corresponding to each of the clothing sample images by using the same three-dimensional model generation network includes: If the clothing sample image is a target segmentation part segmented from the original resolution image, using the three-dimensional model generation network to generate the work clothing three-dimensional model of the clothing sample image.

6. The work clothing dressing detection method according to claim 1, characterized in that The comparing the similarity between the clothing three-dimensional model and each of the work clothing three-dimensional models to obtain a comparison result includes: Outputting each of the work clothing three-dimensional models and the model information of each of the work clothing three-dimensional models; the model information includes at least one of model score information, clothing type information, positive and negative sample information, and positive and negative angle information; In response to the user's selection operation, a target work uniform three-dimensional model is determined from all the work uniform three-dimensional models; The clothing three-dimensional model is respectively compared with each of the target work uniform three-dimensional models for similarity to obtain a comparison result.

7. The work clothing dressing detection method according to claim 1, characterized in that After determining whether the person to be detected is correctly dressed based on the comparison result, the method further includes: If it is determined that the person to be detected is not correctly dressed, the dressed image is used as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

8. The work uniform dressing detection method according to claim 7, wherein, The step of using the dressed image as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear if it is determined that the person to be detected is not correctly dressed includes: If the sample type of the clothing sample image with the highest similarity in the comparison result is a negative sample, it is determined that the person to be detected is not correctly dressed; The dressed image is input into a work uniform confidence calculation model to calculate the confidence of the dressed image being a work uniform, and a confidence prediction result output by the work uniform confidence calculation model is obtained; If the confidence prediction result is greater than a preset threshold, the dressed image is used as a negative sample image to update the clothing sample image corresponding to the clothing that the person to be detected should wear.

9. The work clothing dressing detection method according to any one of claims 1 to 8, characterized in that After determining whether the person to be detected is correctly dressed based on the comparison result, the method further includes: If the dressed image meets the alarm condition, a target box reminder mark is superimposed on the clothing picture part to obtain an alarm image; Output the alarm image.

10. A work uniform wearing detection device, characterized in that, Including: A processor, a memory, and a computer program stored in the memory, and the steps of the work uniform dressing detection method according to any one of claims 1 to 9 are implemented when the computer program is run by the processor.

Images