Multispectral human body recognition control method, recognition system, storage medium and equipment

Abstract

The invention provides a multispectral human body recognition control method, a recognition system, a computer readable storage medium and equipment. The multispectral human body recognition control method comprises the steps that multiple pieces of first multispectral image data are obtained, each piece of first multispectral image data comprises image data of human skin and image data of non-human skin at the same time, the multiple pieces of first multispectral image data are labeled so as to obtain a human skin sample and a non-human skin sample, and the human skin sample and the non-human skin sample are compared; training the labeled human body skin sample and the non-human body skin sample by adopting a machine learning method to obtain a pixel-level multispectral recognition model, inputting collected second multispectral image data of the target object into the pixel-level multispectral recognition model to obtain a human body skin binary image corresponding to the second multispectral image data, and outputting the human body skin binary image. And whether the human body exists in the target object is judged according to the human body skin binary image, so that the recognition precision is improved, and the time required for recognition is shortened.

Description

Technical Field

[0001] The present application relates to the technical field of human identification, and in particular to a multi-spectral human identification control method, an identification system, a computer-readable storage medium and a device. Background Art

[0002] In some scenarios, it is necessary to detect and distinguish people and objects in the target area. If someone enters the target area, the equipment needs to be quickly controlled to stop or avoid to prevent people from getting hurt. Currently, industrial cameras are generally installed to obtain real-time images in the target area, and deep learning models are used to distinguish people. However, due to the complexity of deep learning models, many parameters, and large amount of calculations, recognition requires a long reaction time. Summary of the invention

[0003] Embodiments of the present application provide a multi-spectral human body recognition control method, a recognition system, a computer-readable storage medium, and a device.

[0004] In a first aspect, an embodiment of the present application provides a multi-spectral human body recognition control method, comprising:

[0005] Acquire a plurality of first multispectral image data, each of the first multispectral image data simultaneously including image data of human skin and image data of non-human skin;

[0006] Annotate the plurality of the first multispectral image data to obtain human skin samples and non-human skin samples;

[0007] The labeled human skin samples and non-human skin samples are trained using machine learning methods to obtain a pixel-level multispectral recognition model;

[0008] Inputting the collected second multispectral image data of the target object into a pixel-level multispectral recognition model to obtain a human skin binary image corresponding to the second multispectral image data;

[0009] It is determined whether there is a human body in the target object according to the human skin binary image.

[0010] In one embodiment, the step of acquiring a plurality of first multispectral image data comprises:

[0011] Multispectral image data including both human skin image data and non-human skin image data are collected respectively in multiple scenes to obtain multiple first multispectral image data.

[0012] In one embodiment, the multiple scenes include at least one of different time, different location and different ambient lighting.

[0013] In one embodiment, the step of labeling the plurality of first multispectral image data to obtain human skin samples and non-human skin samples comprises:

[0014] Respectively extracting a first pixel set and a second pixel set from the plurality of the first multispectral image data, wherein the first pixel set includes a plurality of first pixels, and the second pixel set includes a plurality of second pixels;

[0015] Each of the first pixels is labeled as a human skin pixel, and each of the second pixels is labeled as a non-human skin pixel, so as to obtain a human skin sample and a non-human skin sample.

[0016] In one embodiment, the step of judging whether there is a human body in the target object according to the human skin binary image comprises:

[0017] When the human skin binary image includes a human skin pixel, determining the number of other human skin pixels around the human skin pixel by a region growing algorithm;

[0018] When the number of other human skin pixels meets a preset value, it is confirmed that a human body exists in the target object.

[0019] In one embodiment, the machine learning method includes Logistic regression, support vector machine or random forest, BP neural network.

[0020] In one embodiment, after determining whether there is a human body in the target object according to the human skin binary image, the method further includes the following steps:

[0021] When it is determined that a human body exists in the target object, the control device performs a first operation;

[0022] When it is determined that there is no human body in the target object, the control device performs a second operation.

[0023] In a second aspect, an embodiment of the present application provides a multi-spectral human body recognition system, using the multi-spectral human body recognition control method as described above, including:

[0024] The multi-spectral human body recognition system comprises:

[0025] A multispectral image acquisition module, used for acquiring the first multispectral image data and the second multispectral image data;

[0026] A multispectral image annotation module, used for annotating the first multispectral image data;

[0027] A model building module is used to train labeled human skin samples and non-human skin samples using a machine learning method to obtain a pixel-level multispectral recognition model;

[0028] The recognition and judgment module is used to input the second multispectral image data of the collected target object into the pixel-level multispectral recognition model, and is suitable for performing recognition and judgment based on the human skin binary image output by the pixel-level multispectral recognition model to determine whether there is a human body in the target object.

[0029] According to the computer-readable storage medium of the third aspect embodiment of the present application, the computer-readable storage medium stores a computer program, and when the computer program is processed and executed, the steps of the multi-spectral human body recognition control method as described above are implemented.

[0030] According to the fourth aspect of the present application, the device includes a control device, which is used to execute the multi-spectral human body recognition control method as described above.

[0031] In an embodiment, the device comprises a vehicle or a processing machine.

[0032] Beneficial effects of the embodiments of the present application:

[0033] In an embodiment of the present application, by annotating and training a plurality of first multispectral image data, the pixel-level multispectral recognition model can learn more diverse features. This diversity makes the pixel-level multispectral recognition model more robust in the face of complex environments (such as illumination changes, occlusion, noise, etc.). In addition, the pixel-level multispectral recognition model obtained by machine learning training can achieve a fast and automated recognition process. When new multispectral image data is input, the model can quickly output the corresponding human skin binary map, which is convenient for judging whether there is a human body in the target object according to the human skin binary map. Because the pixel-level multispectral recognition model can classify and identify each pixel in the image, rather than just making a rough judgment on the entire image. This fine-grained recognition method helps to improve the overall recognition accuracy. In addition, the method uses a machine learning method to automatically train the labeled human skin samples and non-human skin samples, and the pixel-level multispectral recognition model can be obtained without manual intervention. This automated process greatly improves processing efficiency and accuracy and reduces the time required for recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1is a flow chart of a multi-spectral human body recognition control method provided by the first embodiment of the present application;

[0036] Figure 2 yes Figure 1 The specific flow chart of step S20 is shown;

[0037] Figure 3 yes Figure 1 The specific flow chart of step S50 is shown;

[0038] Figure 4 is a flow chart of a multi-spectral human body recognition control method provided by a second embodiment of the present application;

[0039] Figure 5 This is a schematic diagram of the structure of a control device for a hardware operating environment involved in an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application. In addition, it should be understood that the specific implementation methods described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In the present application, unless otherwise stated, the directional words such as "upper" and "lower" used generally refer to the upper and lower parts of the device in actual use or working state, specifically the drawing direction in the accompanying drawings; while "inside" and "outside" refer to the outline of the device.

[0041] Combine the following Figures 1 to 5 The present invention describes a multi-spectral human body recognition control method, a recognition system, a computer-readable storage medium and a device.

[0042] According to the embodiment of the first aspect of the present application, Figure 1 , the multi-spectral human body recognition control method includes:

[0043] Step S10: Acquire a plurality of first multispectral image data, each of which contains image data of human skin and image data of non-human skin.

[0044] It should be noted that multispectral imaging technology uses the spectral reflectance characteristics of objects at different wavelengths to obtain image data of multiple spectral channels through an optical system. These spectral channels usually cover multiple bands such as visible light, infrared light, and ultraviolet light, so that the subtle differences of objects under different spectra can be captured. Multispectral images have a high spectral resolution, that is, they can distinguish the subtle differences of objects in different spectral bands. This high resolution enables multispectral images to more accurately reflect the spectral characteristics of objects, including the characteristics of human skin. When acquiring the first multispectral image data, the imaging system simultaneously captures the image data of human skin and non-human skin. These data contain rich spectral information, which provides a basis for subsequent processing and analysis. Since multispectral images can capture the unique characteristics of human skin under different spectra, compared with single spectrum image data, multispectral images can more accurately identify human skin and reduce the possibility of misjudgment and missed judgment. Multispectral imaging technology can reduce the impact of environmental factors on recognition results. For example, factors such as lighting changes, shadows, and background complexity may cause significant interference to the recognition results in a single spectrum image, but in a multispectral image, due to the different sensitivity of different spectrums to the environment, by comprehensively analyzing the data of multiple spectrums, it is possible to more effectively resist external interference and improve the robustness of recognition. Each first multispectral image data contains both human skin and non-human skin image data, which enables subsequent processing and analysis to be based on more comprehensive information.

[0045] Step S20: annotate the plurality of first multispectral image data to obtain human skin samples and non-human skin samples.

[0046] It should be noted that, in one embodiment, the plurality of first multispectral image data may be annotated manually. In other embodiments, the annotation may be performed automatically by an annotation tool. Common annotation tools include image annotation software, deep learning frameworks, etc. These tools generally provide rich annotation functions and user interfaces, and support pixel-level or region-level annotation of multispectral images.

[0047] When annotation is required, the annotation process may involve classifying or annotating each pixel, or it may involve marking specific areas in the image. The annotator needs to accurately classify each pixel or area into the corresponding category based on the spectral characteristics and category definition of the image. During the annotation process, the annotation quality needs to be controlled. This can be done by consistency checks between multiple annotators, verification of annotation results, etc. to ensure the accuracy and consistency of the annotation results. Quality control is an important part of ensuring the quality of the annotation data.

[0048] In addition, the annotated multispectral image data provides a large number of high-quality training samples for machine learning algorithms. These annotated data contain rich spectral information and accurate category labels, which help machine learning models learn more accurate feature representations, thereby improving the recognition accuracy and generalization ability of the model. The annotated multispectral image data can help the model capture the subtle differences between different samples and achieve more accurate recognition.

[0049] Step S30: training the labeled human skin samples and non-human skin samples using a machine learning method to obtain a pixel-level multispectral recognition model.

[0050] In this step, a large amount of first multispectral image data is obtained, which usually contains spectral information of different bands, such as visible light, near infrared, short-wave infrared, etc. The collected multiple first multispectral image data are annotated at the pixel level, that is, each pixel point is classified or identified, and the target category or feature to which it belongs is marked. Using the characteristics of multispectral images, useful spectral features are extracted from each image. These features may include spectral characteristics such as reflectivity and absorptivity in different bands. The feature extraction process can be achieved through machine learning methods. The annotated multispectral image data is divided into a training set and a test set, and the model is trained using the training set data. During the training process, the model will learn how to map the input multispectral image features to the corresponding pixel-level labels. The trained model is evaluated using the test set data to check its recognition accuracy and generalization ability. The model is optimized according to the evaluation results, and finally a pixel-level multispectral recognition model that meets the requirements is obtained.

[0051] It should be noted that optimizing the model may include adjusting model parameters, replacing algorithms, or improving feature extraction methods, etc. Specifically, the present application does not limit the specific method of optimizing the pixel-level multispectral recognition model.

[0052] In addition, machine learning methods can automatically learn the complex features and patterns in multispectral images, so as to more accurately identify the targets or features in the images. Compared with traditional physical model-based methods, machine learning methods usually have higher recognition accuracy and better generalization ability. Machine learning models can show strong robustness when processing complex and changeable multispectral image data. Even in the face of adverse factors such as noise, occlusion, and illumination changes, the model can still maintain high recognition performance. Through pixel-level annotation and training, the model can classify or identify each pixel in the image and achieve high-precision spatial resolution. This is especially important for application scenarios that require precise spatial information (such as agricultural remote sensing, environmental monitoring, etc.). The continuous development of modern machine learning algorithms and hardware has greatly increased the speed of processing large-scale multispectral image data. This helps to achieve real-time or quasi-real-time image analysis and recognition tasks, and can also reduce the time required for recognition.

[0053] It should be noted that machine learning methods include logistic regression, support vector machine, random forest or BP neural network.

[0054] Step S40: input the collected second multispectral image data of the target object into a pixel-level multispectral recognition model to obtain a human skin binary image corresponding to the second multispectral image data.

[0055] In this step, first, a second multispectral image data of the target object is collected by a multispectral camera or other multispectral imaging equipment. These data contain the reflection or radiation information of the target object in different spectral bands, usually covering multiple bands such as visible light, near infrared, and short-wave infrared.

[0056] The pixel-level multispectral recognition model is trained in the early stage through a large amount of annotated multispectral image data. The model can learn and recognize the characteristics of human skin under different spectra, and has the ability to apply these characteristics to new image data. When the second multispectral image data is input into the model, the model will extract features for each pixel in the image. These features usually include information such as reflectivity, brightness, and color of pixels in different spectral bands. The pixel-level multispectral recognition model will determine whether each pixel belongs to human skin based on these features. Finally, the pixel-level multispectral recognition model will output a binary image of human skin. In this binary image, pixels belonging to human skin are marked as 1 (or white), and pixels that are not human skin are marked as 0 (or black). In this way, accurate segmentation of the human skin area in the target object is achieved.

[0057] It should be noted that the collected second multispectral image data is used as input, and these data contain information about the target object in different spectral bands, such as visible light, near infrared, short-wave infrared, etc. The input data is usually preprocessed, such as denoising, normalization, etc., to ensure that the pixel-level multispectral recognition model can process efficiently and accurately.

[0058] Step S50: judging whether there is a human body in the target object according to the human skin binary image.

[0059] When the second multispectral image data of the target object is input into the pixel-level multispectral recognition model, the pixel-level multispectral recognition model will classify or recognize each pixel in the image according to the previously learned rules or parameters, and finally output a human skin binary image. In this step, based on the human skin binary image output by the model, it can be comprehensively judged whether there is a human body in the target object.

[0060] In an embodiment of the present application, by annotating and training a plurality of first multispectral image data, the pixel-level multispectral recognition model can learn more diverse features. This diversity makes the pixel-level multispectral recognition model more robust in the face of complex environments (such as illumination changes, occlusion, noise, etc.). In addition, the pixel-level multispectral recognition model obtained by machine learning training can achieve a fast and automated recognition process. When new multispectral image data is input, the model can quickly output the corresponding human skin binary map, which is convenient for judging whether there is a human body in the target object based on the human skin binary map. Because the pixel-level multispectral recognition model can classify and identify each pixel in the image, rather than just making a rough judgment on the entire image. This fine-grained recognition method helps to improve the overall recognition accuracy. In addition, the method uses a machine learning method to automatically train the labeled human skin samples and non-human skin samples, and the pixel-level multispectral recognition model can be obtained without manual intervention. This automated process greatly improves processing efficiency and accuracy and reduces the time required for recognition.

[0061] Specifically, the step of acquiring a plurality of first multispectral image data in step 10 includes:

[0062] Step 101 : collecting multispectral image data including both human skin image data and non-human skin image data in multiple scenes respectively to obtain multiple first multispectral image data.

[0063] In this step, multispectral image data containing both human skin image data and non-human skin image data are collected in multiple scenes to obtain multiple first multispectral image data, so that the diversity and representativeness of the data can be ensured. This diversity helps the model learn more diverse features and improves the generalization ability and robustness of the pixel-level multispectral recognition model. By collecting the first multispectral image data in multiple scenes, the model can learn more diverse features, thereby improving the recognition accuracy of human skin and non-human skin. This accuracy improvement helps to achieve more accurate human skin detection in complex environments. Since the multiple first multispectral image data come from multiple scenes, the pixel-level multispectral recognition model can show stronger robustness in the face of different scenes. This robustness makes the pixel-level multispectral recognition model more reliable and stable in practical applications.

[0064] It should be noted that different scenes may include different shooting angles, different skin conditions (such as dry, moist, with cosmetics, etc.), different temperatures and different weather, etc. Specifically, in the embodiments of the present application, the multiple scenes include at least one of different time, different location and different ambient lighting. Of course, in other embodiments, different scenes can be set as needed, and the present application does not limit this.

[0065] Reference Figure 2 In one embodiment, step S20 of annotating a plurality of first multispectral image data to obtain human skin samples and non-human skin samples includes:

[0066] Step S201, respectively extracting a first pixel set and a second pixel set from a plurality of first multispectral image data, wherein the first pixel set includes a plurality of first pixels, and the second pixel set includes a plurality of second pixels;

[0067] It should be noted that in a multispectral image, each pixel contains numerical information of multiple spectral bands, which can reflect the characteristics of the pixel under different spectra. By extracting a set of pixels in the image, these pixels can be classified according to their spectral characteristics. The purpose of this step is to extract a plurality of first pixels and a plurality of second pixels from a plurality of first multispectral image data, and the plurality of the plurality of first pixels and the plurality of the plurality of second pixels are both greater than the plurality of the plurality of first multispectral image data. A plurality of first pixels constitute a first pixel set, and a plurality of second pixels constitute a second pixel set. Usually, this step can be completed by writing a specific image processing algorithm or using ready-made image processing software. The algorithm or software will automatically or manually identify and extract the first pixel set and the second pixel set based on the spectral characteristics of the multispectral image.

[0068] Step S202: label each first pixel as a human skin pixel, and label each second pixel as a non-human skin pixel, so as to obtain a human skin sample and a non-human skin sample.

[0069] In this step, the first pixel set (including multiple first pixels) is assumed to be a pixel set representing a human skin area, while the second pixel set (including multiple second pixels) represents a pixel set in a non-human skin area. Using computer vision and image processing technology, feature analysis is performed on the extracted pixel set. These features may include color, texture, spectral reflectance, etc., which may be different in different types of skin and non-skin areas. Based on these features, each pixel in the first pixel set is labeled as a human skin pixel, and each pixel in the second pixel set is labeled as a non-human skin pixel.

[0070] It should be noted that, in one embodiment, each first pixel can be labeled as a human skin pixel, and each second pixel can be labeled as a non-human skin pixel by manual labeling. In other embodiments, each first pixel can be labeled as a human skin pixel, and each second pixel can be labeled as a non-human skin pixel by labeling software. Specifically, the present application does not limit the specific labeling method of the first pixel and the second pixel. In addition, the human skin sample is a set of pixels labeled as human skin, and the non-human skin sample is a set of pixels labeled as non-human skin.

[0071] Specifically, the first pixel and the second pixel adopt a binary classification standard. In machine learning, the purpose of binary classification is to divide the input data into two categories. In this step, the extracted first pixel and the second pixel are labeled as two categories of "human skin pixel" and "non-human skin pixel", respectively. The labeling process can be completed by manual labeling or using an automatic labeling algorithm. Manual labeling requires the annotator to classify each pixel based on professional knowledge and experience. Automatic labeling relies on a trained classification model that can automatically determine the category of the pixel. Through pixel extraction and binary classification labeling, it can be ensured that the data used for subsequent training or analysis is accurate and reliable. This helps to improve the performance and accuracy of the pixel-level multispectral recognition model.

[0072] In one embodiment, referring to Figure 3 Step S50 of judging whether there is a human body in the target object according to the human skin binary image comprises:

[0073] Step S501: When a human skin binary image includes a human skin pixel, the number of other human skin pixels around the human skin pixel is determined by a region growing algorithm.

[0074] It should be noted that whether the human skin binary image includes human skin pixels can be determined by traversing the annotation set and checking the category label of each annotation to determine whether there are human skin pixels.

[0075] In this step, first, at least one human skin pixel is identified in the human skin binary image as an initial seed point. These seed points are the starting point of the subsequent region growing process. Starting from the initial seed point, the region growing algorithm will check its adjacent pixels. If the adjacent pixels meet certain similarity criteria (here they belong to human skin pixels), these pixels are added to the current region and continue to grow outward as new seed points. This process is iterated until no more adjacent pixels meet the similarity criteria. After the region growing is completed, the number of other human skin pixels around the human skin pixel as the seed point is calculated.

[0076] It should be noted that the region growing algorithm can more accurately identify and connect skin pixels belonging to the same human body, reducing the possibility of misjudgment and missed judgment. This helps to improve the accuracy of judgment. The region growing algorithm has a certain inhibitory effect on noise and interference in the image, and can reduce the impact of these factors on the judgment results to a certain extent. This makes the method more robust in practical applications. The region growing algorithm is relatively simple and easy to implement on a computer. In application scenarios that require automatic identification of human bodies (such as intelligent monitoring, intelligent control, human-computer interaction, etc.), quickly and accurately determining whether there is a human body in the target object is the key to improving user experience. This method can significantly improve the efficiency and accuracy of judgment, thereby improving user satisfaction and experience and reducing user waiting time.

[0077] Step S502: When the number of other human skin pixels meets a preset value, it is confirmed that there is a human body in the target object.

[0078] In this step, the number of other human skin pixels around the human skin pixel of the seed point is compared with the preset value. If the number of other human skin pixels around the seed point meets or exceeds the preset value, it is considered that there is a human body in the target object; otherwise, it is considered that there is no human body in the target object or the human body part is too small to be recognized.

[0079] It should be noted that this step is based on a preset threshold (preset value), which represents the minimum number of human skin pixels required for an area to be considered as a "human body".

[0080] Specifically, in the above steps, by combining the human skin binary image and the region growing algorithm, the method can more accurately identify the skin area in the image, reducing the possibility of misjudgment and missed judgment. By first detecting the existence of human skin pixels and then performing a detailed region growing analysis on possible skin areas, the method can reduce unnecessary calculations and improve processing speed while ensuring accuracy.

[0081] Reference Figure 4 In one embodiment, after step S50 determines whether there is a human body in the target object according to the human skin binary image, the following steps are also included:

[0082] Step S60: When it is determined that there is a human body in the target object, the control device performs a first operation;

[0083] Step S70: When it is determined that there is no human body in the target object, the control device performs a second operation.

[0084] In the above steps, the corresponding operation is performed directly according to the judgment result, and the device can respond to user needs or environmental changes more quickly. This instant feedback mechanism improves the overall responsiveness of the system and user experience. These two steps realize the automatic control of the device, and the corresponding operation can be automatically performed according to the judgment result without human intervention. This not only improves processing efficiency, but also reduces the possibility of human error. By designing different operations for different judgment results, the device can flexibly adjust its behavior according to actual needs.

[0085] It should be noted that the above-mentioned first operation and second operation can be set according to actual conditions. For example, in one embodiment, when the device is a processing machine, when it is determined that there is a human body in the target object, the control device performs the first operation specifically to stop the processing machine from working to avoid the processing machine from accidentally injuring the operator. When it is determined that there is no human body in the target object, the control device performs the second operation specifically to control the processing machine to continue working. In another embodiment, for example, in military use, the device can be a weapon device, and when it is determined that there is a human body in the target object (determined as a living target of the enemy), the control device performs the first operation specifically to control the weapon device to attack. When it is determined that there is no human body in the target object (determined as a non-living target), the control device performs the second operation specifically to control the weapon device to stop attacking. In other embodiments, when the device is an unmanned vehicle, when it is determined that there is a human body in the target object, the control device performs the first operation specifically to control the unmanned vehicle to avoid. When it is determined that there is no human body in the target object, the control device performs the second operation specifically to control the unmanned vehicle to continue driving.

[0086] According to an embodiment of the second aspect of the present application, a multispectral human body recognition system is provided. The multispectral human body recognition system uses the multispectral human body recognition control method as described above. The multispectral human body recognition system includes a multispectral image acquisition module, a multispectral image annotation module, a model construction module, a model construction module and a recognition judgment module.

[0087] The multispectral image acquisition module is used to acquire first multispectral image data and second multispectral image data.

[0088] It should be noted that the principle of the multispectral image acquisition module is based on the basic principles of photophysics and spectroscopy, and adopts a multi-band imaging method. The module can simultaneously obtain the spectral information of the target object in different bands by integrating spectral sensors or spectrometers of multiple discrete bands. These spectral information reflects the reflection, radiation or emission characteristics of the target substance at different wavelengths, thereby revealing detailed information such as the chemical composition, optical properties, and physiological state of the target substance. When collecting the first multispectral image data and the second multispectral image data, the multispectral image acquisition module will irradiate light of different wavelengths onto the target object according to the preset band range using a special light source device and a corresponding excitation filter. Subsequently, the reflected light or emitted light of the target object in different bands is collected by imaging devices such as high-resolution CCD or digital cameras to form the first multispectral image data or the second multispectral image data. These data can be used to identify, classify, and characterize the characteristics of the target object in subsequent processing and analysis.

[0089] In addition, the multispectral image acquisition module has a higher detection efficiency. It can simultaneously obtain the spectral information of the target in different bands in a short time, greatly shortening the detection time.

[0090] The multispectral image annotation module is used to annotate the first multispectral image data.

[0091] It should be noted that the principle of the multispectral image annotation module involves image processing and computer vision technology, and is intended to accurately annotate multispectral image data. This process usually includes steps such as data preprocessing, annotation category definition, annotation tool selection, pixel-level annotation, annotation quality control, and export of annotation data sets. Data preprocessing is specifically manifested as: preprocessing the first multispectral image data, including denoising, correction, and enhancement operations to improve image quality and visualization effects for subsequent annotation.

[0092] In addition, in the first multispectral image data, since each pixel contains the values ​​of multiple spectral bands, pixel-level annotation is usually required. This requires the annotator to classify or annotate each pixel to ensure the accuracy of the annotation results. Pixel-level annotation can be achieved by manual drawing or using an automatic segmentation algorithm. During the annotation process, the annotation quality needs to be controlled. This can be done by consistency checks between multiple annotators, verification of annotation results, etc. to ensure the accuracy and consistency of the annotation results. In addition, machine learning algorithms can be used to automatically evaluate and optimize the annotation results. After the annotation is completed, the annotation results are exported as a data set in a standard format.

[0093] The model building module is used to train labeled human skin samples and non-human skin samples using machine learning methods to obtain a pixel-level multispectral recognition model.

[0094] It should be noted that the labeled human skin samples and non-human skin samples are used as training data sets. These data sets have been preprocessed and annotated, and contain rich spectral information and pixel-level labels. Machine learning algorithms need to extract useful features from these multispectral images. These features may include pixel values, texture information, shape information, etc. in different spectral bands. The purpose of feature extraction is to convert the original image data into a form that the algorithm can understand and process. Select a suitable machine learning model based on the task requirements and data characteristics. For pixel-level multispectral recognition tasks, these machine learning models are able to process image data and learn the mapping relationship from input images to output labels. The selected model is trained using the training data set. During the training process, the algorithm continuously adjusts the model parameters to minimize the prediction error. This usually involves an iterative process of forward propagation (calculating prediction values) and back propagation (adjusting parameters based on prediction errors). After the training is completed, the model is evaluated using an independent test data set. The model is further optimized based on the evaluation results to finally obtain a pixel-level multispectral recognition model.

[0095] The recognition and judgment module is used to input the collected second multispectral image data of the target object into the pixel-level multispectral recognition model, and is suitable for performing recognition and judgment based on the human skin binary image output by the pixel-level multispectral recognition model to determine whether there is a human body in the target object.

[0096] It should be noted that the collected second multispectral image data is input into a pre-trained pixel-level multispectral recognition model. The pixel-level multispectral recognition model is capable of understanding and analyzing image features under different spectral bands. The pixel-level multispectral recognition model will extract features from the input image data, especially for the spectral features unique to human skin (such as reflectivity, absorptivity, etc.). These features are usually related to components such as moisture, pigment, collagen, etc. in the skin. Subsequently, the pixel-level multispectral recognition model will compare the extracted features with the features of human skin samples and non-human skin samples pre-stored and annotated in the database. Through comparison and analysis, the pixel-level multispectral recognition model can determine whether the second multispectral image data contains spectral features unique to human skin, and output the corresponding human skin binary map accordingly. The recognition and judgment module will determine whether there is a human body in the target object based on the human skin binary map.

[0097] In an embodiment of the present application, the multispectral human recognition system is composed of four main parts: a multispectral image acquisition module, a multispectral image annotation module, a model construction module, and a recognition and judgment module. Each module performs its own duties and jointly realizes the accurate recognition of the target object (such as the human body). The multispectral image acquisition module is responsible for collecting the first multispectral image data and the second multispectral image data. These data cover multiple spectral bands such as visible light, near infrared, and short-wave infrared, providing a rich information basis for subsequent recognition. Through multispectral imaging technology, image details under different bands can be captured, providing more comprehensive data support for the recognition model. The multispectral image annotation module annotates the first multispectral image data collected to provide accurate label information for subsequent training. The accuracy of the annotation directly affects the performance of the training model. Through fine annotation, it can be ensured that the model learns the correct feature representation. The model construction module uses a machine learning method to train the annotated human skin samples and non-human skin samples to construct a pixel-level multispectral recognition model. Through training, the model can learn the spectral feature differences between skin samples and non-skin samples, and has the ability to accurately classify new samples. The recognition and judgment module inputs the collected second multispectral image data of the target object into the trained pixel-level multispectral recognition model, and determines whether there is a human body in the target object according to the human skin binary image output by the model. The multispectral human body recognition system can recognize the target object in real time and output accurate recognition results. It has fast recognition speed and high accuracy, which can meet the needs of practical applications. Due to the use of multispectral image data and machine learning technology, the system can achieve high-precision recognition. Multispectral image data provides rich spectral information, allowing the model to learn more sophisticated feature representations; while machine learning technology can automatically extract useful information from the data and build a high-performance recognition model. Compared with single-spectrum image recognition technology, this system has stronger anti-interference ability. Multispectral image data can capture image information in different bands, so that the model can better cope with the influence of complex environmental factors such as lighting changes and shadow interference. The system can provide users with more accurate and personalized services.

[0098] According to an embodiment of the third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is processed and executed, the multi-spectral human body recognition control method as described above is implemented, and the multi-spectral human body recognition control method includes:

[0099] Acquire a plurality of first multispectral image data, each of the first multispectral image data simultaneously including image data of human skin and image data of non-human skin;

[0100] Annotating the plurality of first multispectral image data to obtain human skin samples and non-human skin samples;

[0101] The labeled human skin samples and non-human skin samples are trained using machine learning methods to obtain a pixel-level multispectral recognition model;

[0102] Inputting the collected second multispectral image data of the target object into a pixel-level multispectral recognition model to obtain a human skin binary image corresponding to the second multispectral image data;

[0103] Determine whether there is a human body in the target object based on the human skin binary image.

[0104] According to an embodiment of the fourth aspect of the present application, a device is provided, the device comprising a control device, the control device being used to execute the multi-spectral human body recognition control method as described above.

[0105] See also Figure 5 , Figure 5 It is a schematic diagram of the structure of a control device for a hardware operating environment involved in an embodiment of the present invention.

[0106] like Figure 5 As shown, the control device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0107] Those skilled in the art will understand that Figure 5 The structure shown in the figure does not constitute a limitation on the control device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0108] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a multi-spectral human body recognition control program.

[0109] The device embodiments described above are merely illustrative, wherein 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, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0110] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0111] It should be noted that there are many types of equipment, for example, the equipment may include medical equipment, weapon devices, etc. Specifically, in the embodiments of the present application, the equipment includes a vehicle or a processing machine. In addition, the type of equipment can be further set as needed, and the present application does not limit this.

[0112] The embodiments of the present application are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A multi-spectral human body recognition control method, characterized in that: include: Acquire a plurality of first multispectral image data, each of the first multispectral image data simultaneously including image data of human skin and image data of non-human skin; Annotate the plurality of the first multispectral image data to obtain human skin samples and non-human skin samples; Using a machine learning method to train the labeled human skin samples and the non-human skin samples to obtain a pixel-level multispectral recognition model; Inputting the collected second multispectral image data of the target object into a pixel-level multispectral recognition model to obtain a human skin binary image corresponding to the second multispectral image data; It is determined whether there is a human body in the target object according to the human skin binary image.

2. The multi-spectral human body recognition control method according to claim 1, characterized in that: The step of acquiring a plurality of first multispectral image data comprises: Multispectral image data including both human skin image data and non-human skin image data are collected respectively in multiple scenes to obtain multiple first multispectral image data.

3. The multi-spectral human body recognition control method according to claim 2, characterized in that: The multiple scenes include at least one of different time, different location and different ambient lighting.

4. The multi-spectral human body recognition control method according to claim 1, characterized in that: The step of labeling the plurality of the first multispectral image data to obtain human skin samples and non-human skin samples comprises: Respectively extracting a first pixel set and a second pixel set from the plurality of the first multispectral image data, wherein the first pixel set includes a plurality of first pixels, and the second pixel set includes a plurality of second pixels; Each of the first pixels is labeled as a human skin pixel, and each of the second pixels is labeled as a non-human skin pixel, so as to obtain a human skin sample and a non-human skin sample.

5. The multi-spectral human body recognition control method according to claim 1, characterized in that: The step of judging whether there is a human body in the target object according to the human skin binary image comprises: When the human skin binary image includes a human skin pixel, determining the number of other human skin pixels around the human skin pixel by a region growing algorithm; When the number of other human skin pixels meets a preset value, it is confirmed that a human body exists in the target object.

6. The multi-spectral human body recognition control method according to claim 1, characterized in that: The machine learning method includes Logistic regression, support vector machine, random forest or BP neural network.

7. The multi-spectral human body recognition control method according to any one of claims 1 to 6, characterized in that: After judging whether there is a human body in the target object according to the human skin binary image, the following steps are also included: When it is determined that a human body exists in the target object, the control device performs a first operation; When it is determined that there is no human body in the target object, the control device performs a second operation.

8. A multi-spectral human body recognition system, using the multi-spectral human body recognition control method according to any one of claims 1 to 7, characterized in that: The multi-spectral human body recognition system comprises: A multispectral image acquisition module, used for acquiring the first multispectral image data and the second multispectral image data; A multispectral image annotation module, used for annotating the first multispectral image data; A model building module, used to train the labeled human skin samples and the non-human skin samples using a machine learning method to obtain a pixel-level multispectral recognition model; The recognition and judgment module is used to input the second multispectral image data of the collected target object into the pixel-level multispectral recognition model, and is suitable for performing recognition and judgment based on the human skin binary image output by the pixel-level multispectral recognition model to determine whether there is a human body in the target object.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is processed and executed, the steps of the multi-spectral human body recognition control method according to any one of claims 1 to 7 are implemented.

10. A device, characterized in that: It comprises a control device, wherein the control device is used to execute the multi-spectral human body recognition control method as described in any one of claims 1 to 7.

11. The device according to claim 10, characterized in that The equipment comprises a vehicle or a processing machine.

Images