A method for indoor human body spatial positioning and posture contour recognition

Abstract

A method for indoor human spatial positioning and posture contour recognition includes a recognition system installed indoors, including a ceiling hotspot millimeter-wave radar and a detection millimeter-wave radar, as well as detector groups on different walls. The detector group includes a detection millimeter-wave radar and a visual camera. The 6-degree-of-freedom detection values obtained by the detection millimeter-wave radar and the visual camera are weighted and averaged based on their credibility to obtain a 6-degree-of-freedom result for the detector group. The detection results of the detector group and the ceiling detection millimeter-wave radar are then summed and averaged to ultimately obtain a 6-degree-of-freedom result for the target person in the world coordinate system. The present invention utilizes relatively low-cost detectors and integrates sensors from multiple angles and different types to achieve accurate indoor positioning and human posture recognition, avoiding errors caused by occlusion and other issues. Real-time detection facilitates prediction of human posture trends and can be used for monitoring behavioral characteristics. It is also suitable for collecting information on diseases exhibiting behavioral characteristics.

Description

Technical Field

[0001] The present invention belongs to the field of sensing detection technology, and relates to the detection of indoor human posture by millimeter wave radar and vision, which is a method for indoor human spatial positioning and posture contour recognition. Background Art

[0002] Currently, most of the millimeter wave spatial positioning technology focuses on the indoor trajectory of people, such as using the millimeter wave TDOA algorithm and Kalman filter to perform indoor static and dynamic positioning of human bodies [1], research on the positioning trajectory of indoor pedestrians [2], and a human motion trajectory detection system based on FMCW radar [3]. However, at present, millimeter wave radar is mainly used to locate the trajectory of people, and it is not possible to locate the posture of the human body. On the other hand, there are also millimeter wave radar human body three-dimensional imaging technologies, such as using millimeter wave technology to perform three-dimensional imaging modeling of the human body during security inspections [4]. However, this method requires the three-dimensional reconstruction to be completed at a fixed point and cannot be completed in real time while the human body is moving. This application proposes a modeling method for real-time positioning and real-time calculation of the real-time posture profile of the target person indoors.

[0003] This application uses indoor spaces as the primary scenario and employs a multi-sensor fusion approach, utilizing visual cameras and millimeter-wave radar sensors. These sensors combine to detect a target person from multiple angles, achieving relatively accurate three-dimensional human body contour modeling at a low cost. The millimeter-wave radar human posture recognition system, modeled after this application, ultimately achieves real-time human posture contour recognition.

[0004] References

[0005] [1] Li Fangxin, Tu Rui, Han Junqiang, Zhang Yin, Hong Ju. Indoor positioning algorithm based on 5G millimeter wave arrival time difference [J]. Global Positioning System, 2021, 46(02): 1-6.

[0006] [2] Yu Dengbo. Behavior pattern recognition and analysis based on indoor pedestrian positioning trajectory[D]. Wuhan University, 2019.

[0007] [3] Yin Huibin, Xu Zhimeng. Human motion trajectory detection system based on FMCW radar[J]. Sensors and Microsystems, 2020, 39(09): 116-118.

[0008] [4]Xie Pengfei. Millimeter wave human 3D imaging and target detection[D]. Xidian University, 2019. Summary of the Invention

[0009] The problem to be solved by the present invention is: how to detect the spatial position and posture of a human body indoors in real time while taking into account both detection accuracy and detection cost.

[0010] The technical solution of the present invention is: a method for indoor human spatial positioning and posture contour recognition, wherein a recognition system is set up indoors, the recognition system is composed of detectors, including a hotspot millimeter-wave radar and a detection millimeter-wave radar set on the ceiling, and a detector group set on different walls, the detector group including a detection millimeter-wave radar and a visual camera;

[0011] First, a hotspot millimeter-wave radar installed on the ceiling of the room detects in real time whether there are people in the room and the spatial range of the people. The detected people are regarded as the target people, and the detector groups installed on different walls and the detection millimeter-wave radar installed on the ceiling are activated according to the detected spatial range to detect the spatial position and posture of the target people in real time. The spatial 6-degree-of-freedom detection values of the key points of the target people and the corresponding visual credibility are obtained from the visual camera, and the spatial 6-degree-of-freedom detection values of the target people and the corresponding radar credibility are obtained from the detection millimeter-wave radar. For each detector group, the 6-degree-of-freedom detection values obtained by the detection millimeter-wave radar and the visual camera are weighted and averaged according to the credibility of the two to obtain the 6-degree-of-freedom result of the detector group. The detection results of the detector groups on different walls and the detection millimeter-wave radar on the ceiling are then combined, and the 6-degree-of-freedom results of each detection are summed and averaged. Finally, the 6-degree-of-freedom data results of the target people in the world coordinate system, including the position and posture angles, are obtained. The 6 degrees of freedom refer to the spatial position and posture of the human body determined by the 3D coordinates of the three key points of the target person's head and shoulders, including the 3D position (x, y, z) and the rotation angles (α, β, γ) of the spatial position.

[0012] This invention uses real-time spatial positioning of indoor human bodies to identify their movement trajectory and outline posture. Based on real-time positioning, millimeter-wave radar and cameras are used to capture human posture at hotspots and generate recognition models, laying the foundation for subsequent real-time spatial and posture positioning of indoor human bodies. This solution can be used to identify the behavioral characteristics of a target person, as well as changes in their posture during movement and rest. This can help identify the presence of early symptoms of behavior-related diseases, or can be used to identify the characteristics of related diseases through behavior.

[0013] The beneficial effects of the present invention are as follows:

[0014] 1. The present invention can use relatively low-cost detectors to achieve real-time and precise positioning of indoor target persons, and perform real-time human posture contour detection and calculation. At present, the average accuracy of high-precision human motion capture is 0.02 degrees, and the delay is less than 20ms. To achieve this type of high-precision human motion capture, it is usually necessary to wear corresponding sensors on the target person. The present invention does not require the target person to wear corresponding sensors, and the accuracy can be close to the accuracy of human motion capture when wearing sensors. In indoor human detection, the detection accuracy of existing millimeter-wave radars or cameras is far lower than the aforementioned high-precision level. Although they can also detect the angular position of the human body, they cannot meet the accuracy and real-time requirements of behavioral feature detection. The present invention can achieve an accuracy of 0.1 degrees and a delay of less than 25ms, which can meet the detection requirements of human behavioral features. At present, the average high-precision motion capture system costs 100,000 to 200,000 yuan per set, and the system cost of the present invention is about 40,000 to 60,000 yuan per set. The present invention can take into account both detection accuracy and detection cost. The present invention can be used to monitor behavioral characteristics, such as preventing falls, and is also suitable for collecting information on diseases with behavioral characteristics, such as Alzheimer's disease. The method of the present invention can be used to collect behavioral information of the target person, which is conducive to early screening based on the early behavioral symptoms of related diseases.

[0015] 2. The present invention integrates sensors of multiple angles and different types to improve the accuracy of indoor positioning and human posture recognition, and avoid errors caused by occlusion and other problems.

[0016] 3. The implementation and layout of the system of the present invention is relatively simple and efficient. When people are active indoors, different groups of detectors can achieve comprehensive recognition and detection. It is preferred to perform a static calibration in the newly arranged room. The subsequent recognition system can dynamically update the positioning and recognition parameters to maintain adaptive and accurate recognition of human dynamic activities.

[0017] 4. The recognition method of the present invention can implement rapid and accurate recognition of the spatial position and posture of the human body, which is conducive to predicting the trend of human posture. For example, based on the result of the posture contour recognition, if the set fall threshold is reached, it is judged that the identified person is about to fall, and an alarm prompt can be issued in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the indoor sensor arrangement of the present invention, including a low-frequency millimeter-wave radar sensor, a millimeter-wave radar sensor, and a visual camera group.

[0019] Figure 2 Schematic diagram of the process of initial calibration of the method of the present invention.

[0020] Figure 3 Schematic diagram of the process of real-time detection of the method of the present invention.

[0021] Figure 4 Schematic diagram of the detection method of the present invention. DETAILED DESCRIPTION

[0022] The implementation of the present invention is described in detail below.

[0023] like Figure 1 As shown, the identification system of the present invention is composed of a series of detectors. 2.4Ghz and 60GHz millimeter-wave radar detectors are set on the top of the room. The 2.4GHz radar is used as a hotspot millimeter-wave radar for initial large-scale detection, and the 60GHz radar is a detection millimeter-wave radar for distance and angle detection facing the hotspot. The 2.4GHz radar has a low frequency, strong penetration, and a wide detection range, and is used for preliminary detection. The 60GHz radar has a high frequency and a relatively small detection range, but has higher accuracy, and is used for subsequent more accurate detection of the target person. Detector groups are set on different walls in the room, and the detector group includes a detection millimeter-wave radar and a visual camera. The detection millimeter-wave radar is also a 60Ghz radar. According to the detection results of the 2.4Ghz radar, the 60Ghz radar on the top and the detector groups on different walls are started to perform directional detection on the hotspot area.

[0024] like Figure 2 As shown, when the present invention is implemented in an indoor setting, it is preferably calibrated first. After arranging the detectors in different planes, a marker cube marked with a special QR code is placed in the center of the floor of the room, and the 2.4GHz and 60GHz radars on the top surface, as well as the 60GHz millimeter-wave radar detectors and visual camera detector groups in the four walls perpendicular to the ground plane are started. The detection results of all detectors on the marker cube are synchronized, and the visual camera and millimeter-wave radar both output 6-degree-of-freedom information of the position and angle of the marker cube to form a vector of [x, y, z, α, β, γ], where x, y, z represent the position information based on the detector coordinates, and α, γ, β represent the angle information corresponding to different coordinate axes in the detector coordinate system. By calculation, the height from the ground to the top surface and the average distance from the detector group to the opposite wall can be obtained. The above data is used as the initialization data of the recognition system. These data can be manually input in advance and automatically recognized through calibration, which is more adaptable to the installed detectors. At the same time, a conversion matrix is formed through the calibration program to convert the position and rotation information in different detector coordinate systems into the world coordinate system. The world coordinate system takes the position of the marker cube as the origin. The coordinate system data and conversion matrix data serve for subsequent data processing. By using the conversion matrix obtained in the calibration step, subsequent data can be directly converted into a unified world coordinate system.

[0025] like Figure 3As shown, during detection and tracking, the approximate position of the human body is first obtained through a 2.4Ghz radar, and then a cylindrical area with a height of 2 meters and a diameter of 1 meter is set as the hotspot area. The 60Ghz radar on the top of the room and the detector group set on the four walls are then used to perform directional detection of the human body in the hotspot area, and the 6-degree-of-freedom spatial data of the target person's head and two shoulders are marked, including three-dimensional position calculation (x, y, z) and rotation calculation (α, β, γ) facing the three coordinate axes. The present invention minimizes the selection of three key points of the head and shoulders, and uses the plane formed by the minimized three non-collinear key points to estimate the posture of the human body, or more precisely, the posture of the upper body of the human body. The posture estimation method is to obtain the normal vector of the plane, and obtain the spatial three-degree-of-freedom angle information of the human body according to the normal vector. The spatial three-degree-of-freedom angle result of the human body posture is used to correct the spatial three-degree-of-freedom angle result of the key point. The corrected key point information is brought into subsequent calculations.

[0026] When measuring the three key points of the target person's head and two shoulders, the BlazePose method can be used, but in the estimated heat map, only the three hot spots of the human head and two shoulders are focused on to reduce the amount of calculation, such as Figure 4 As shown. At the same time, the method of the present invention also increases the credibility of the data of the detection results according to the different characteristics and advantages of the detectors. Each detector needs to dynamically calculate the credibility, and the credibility value is oriented to the 6-degree-of-freedom data, that is, the credibility value on each degree of freedom is different. For a detector group composed of a group of visual detectors and millimeter-wave radar detectors, a weighted average method combined with credibility is used to obtain a unique 6-degree-of-freedom result obtained by the detector group, and then the unique detection results obtained by the detector group of 4 different walls, a total of 4 results and one result of the millimeter-wave radar detection on the top surface are combined with credibility and weighted averaged to finally obtain the 6-degree-of-freedom data results containing the position and angle of the three key points of the target person's head and shoulders in the world coordinate system.

[0027] The detailed steps for calculating the key points of the target person are as follows.

[0028] For a detector group consisting of a visual camera and a millimeter wave detector, calculations are first performed within the detector group. All detectors are started. At this time, it is ensured that there is a detectable human body in the room, that is, the data obtained by each group of detectors is not empty. The data after the first detection of the key points of the human body is obtained. For a certain key point, the 6-DOF signal detection value {P xcamera , P ycamera , P zcamera , P αcamera , P βcamera , P γcamera}, and the corresponding visual credibility T camera, the millimeter wave radar can obtain the target person's 6-DOF detection value {P xradar , P yradar , P zradar , P αradar , P βradar , P γradar}, and the corresponding radar credibility T radar ,The subscripts x, y, z, α, β, and γ represent the six spatial degrees of freedom, where x, y, and z represent position information; α, β, and γ represent angle information.

[0029] All credibility are assigned an initial value, such as 0.1. The initial value is assigned for the convenience of the identification system when performing the first operation to avoid the problem of credibility null value. zcamera and P zradar The value of will change accordingly, and different credibility T is calculated in real time. The z-axis direction detection value P obtained in real time in this group of detectors is zcamera and P zradar To calculate the credibility of the benchmark, calculate the real-time credibility of the detection millimeter wave radar and visual camera:

[0030]

[0031]

[0032] d′ is the average distance from the detector group to the opposite wall plane, which can be obtained by manual input when setting up recognition, or automatically obtained through the calibration method described above.

[0033] The credibility of the present invention is used to judge the distance between the current detector and the target person, according to the vertical distance P of the detector relative to a key point of the target person. zcamera or P zradar , divide the vertical distance by the total distance d', and then subtract the result from the previous step from 1 to obtain the corresponding detector's credibility. This method uses the credibility parameter to enhance the accuracy of the detector group's detection values based on the detector's own detection strengths, thereby improving the precision of the integrated detection values.

[0034] On this basis, in order to reduce the amount of calculation and improve the calculation speed of the recognition system, the present invention also proposes the following two optimization schemes.

[0035] 1) When the visual credibility of a detector T camera Or radar credibility T radarWhen the value is less than 0.5, the 6-DOF value detected by this detector is not included in the calculation of the 6-DOF result of the detector group it belongs to, that is, the value is 0. A confidence level less than 0.5 indicates that the target person is closer to the opposite wall, and the detection results of the detector installed on the opposite wall will be more accurate. In this case, the value of the detector corresponding to the confidence level less than 0.5 is not included in the calculation, thereby reducing the magnitude and scale of the overall recognition system calculation and improving calculation speed. At the same time, since there is no corresponding detector on the bottom surface, the detector on the top surface does not follow this calculation principle.

[0036] 2) For a detector group, if T camera <T radar , then retain T radar and the 6-DOF detection value detected by its corresponding detector, otherwise T is retained camera And the 6-DOF detection value detected by the corresponding detector. That is, when the credibility of the millimeter-wave radar and the visual camera in the detector group is relatively high, one of them is selected as the detection result of the detector group to reduce the amount of calculation.

[0037] The above method is the innovation of the present invention. For different types of detectors, based on their unique characteristics, the present invention sets real-time credibility. According to the real-time detection results of the detector, the detector value with high credibility is selected for calculation as the position and posture of the target person changes. On the one hand, the detection accuracy is improved. On the other hand, the real-time nature of the detection is further guaranteed by optimizing the calculation amount, and the detection delay is small.

[0038] At this time, the integrated result of the detection numerical results of this detector group is P group The calculation formula is as follows:

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] The above method can be used to obtain the integrated results of the detection values of a key point by combining n sets of detectors on n walls. On this basis, the detection data of the millimeter-wave radar on the top surface is added. For the convenience of calculation, all detection results need to be uniformly converted to the world coordinate system using the conversion matrix obtained in the calibration phase. Among them, the position information of the millimeter-wave radar on the top surface is relatively accurate in the world coordinate system, so only the position information of the millimeter-wave radar is used in the calculation. Then P world The calculation formula is:

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] n represents the number of detector groups, which is usually 4 indoors. xgroup_i , P ygroup_i , P zgroup_i , P αgroup_i , P βgroup_i , P γgroup_i} represents the 6-DOF result of the i-th detector group, {P xtop_radar , P ytopradar , P ztop_radar} represents the three-dimensional position detection result of the top surface detection millimeter wave radar, and the final P world ={P xworld , P yworld , P zworld , P αworld , P βworld , P γworld} is the 6-DOF value of the key point after the fusion of heterogeneous sensors on each wall.

[0053] If detection is unsuccessful, meaning all detectors return null values or their reliability falls below the set threshold, indicating they are unreliable, the 2.4GHz millimeter-wave radar detector will be restarted to search for the target person indoors. The search continues at set intervals until the person is found, and then the top-surface millimeter-wave radar and detector group will be tested. During the 2.4GHz radar search, any remaining detectors that remain idle for more than 5 minutes will enter a dormant state until reactivated.

[0054] Furthermore, to ensure the accuracy of detection and recognition results, the present invention performs real-time corrections on the angle values calculated for the key points. Specifically, the original angle estimates for the head and shoulders are corrected based on the calculated angle values of the human body posture. The angle information calculated for the three key points is fed back to the three key points, correcting the original angle estimates for the three key points so that the angle estimates for the key points are synchronized with the angle values of the human body posture.

[0055] Based on the 3D coordinates of three different lines on the head and shoulders, the normal vector to the plane formed by these three points is calculated, along with the rotation angles of this normal vector about the three coordinate axes in the world coordinate system. This provides an estimate of the human pose. Based on this estimate, the angles α, β, and γ relative to the original three key points are corrected using this estimate.

[0056] ΔR jα =R jα -R humanα

[0057] ΔR jβ =R jβ -R humanβ

[0058] ΔR jγ =R jγ -R humanγ

[0059] R j ' α =R jα +ΔR jα

[0060] R j ' β =R jβ +ΔR jβ

[0061] R j ' γ =R jγ +ΔR jγ

[0062] j is the number of the key point, j = 1, 2, 3, R jα 、R jβ 、R jγ is the angle value obtained for key point j, R humanα 、R humanβ 、R humanγ is the rotation angle obtained by the normal vectors of the three key points, R j ' α 、R j ' β 、R j 'γ It represents the corrected angle value of the jth key point. First, find the difference between the estimated human posture angle and the angle calculated by the original key point. Then, use this difference as the angle correction value for the next operation, and continuously correct the angle calculated by the key point.

[0063] In real-time detection, the detector calculates the angle value of the human body in real time, ΔR j Indicates the angle correction value obtained from the last calculation result. According to the timing, when the first calculation is performed, R j The detector directly calculates the key point angle value, not the first calculation R j To obtain the key point angle value from the last calculation. human Based on the 6-DOF vector results of the three key points obtained by the detector, the angular rotation value is obtained by calculating the normal vector of the plane where the three key points are located. j The angle difference obtained in the first calculation is used as the initial value of the corrected parameter. This difference calculation is repeated every time a new human posture estimate is obtained, and the new corrected difference is then brought into the next calculation. This is a key innovation of the present invention. The position information of the three key points is used to form a plane and then calculate the normal vector to obtain a human posture estimate. The angle value in the plane is then used to correct the angle value obtained by the detector to continuously approach the correct angle value. This improves the accuracy of the next human posture estimate.

[0064] In existing technologies, millimeter-wave sensors can detect human motion, but their accuracy is limited and they are easily affected by other objects. Therefore, visual recognition results are introduced as an aid. However, these results can also be affected by occlusion, leading to inaccuracies. Millimeter-wave sensors are superior to visual sensors in measuring distance and angle, as visual sensors use distance and angle estimates. This present invention proposes a method for using different wall detector groups, assigning credibility levels to the results of different detectors. Based on the dynamic credibility, the weights of different sensor detection results in the final result are dynamically adjusted to ensure the accuracy of the overall recognition system's real-time detection and recognition.

Claims

1. A method for indoor human body spatial positioning and posture contour recognition, characterized by An indoor identification system is set up. The identification system consists of detectors, including a hotspot millimeter-wave radar and a detection millimeter-wave radar installed on the ceiling, and a detector group installed on different walls. The detector group includes a detection millimeter-wave radar and a visual camera. First, a hotspot millimeter-wave radar installed on the ceiling of the room detects in real time whether there are people in the room and the spatial range of the people. The detected people are regarded as the target people, and the detector groups installed on different walls and the detection millimeter-wave radar installed on the ceiling are activated according to the detected spatial range to detect the spatial position and posture of the target people in real time. The spatial 6-degree-of-freedom detection values of the key points of the target people and the corresponding visual credibility are obtained from the visual camera, and the spatial 6-degree-of-freedom detection values of the target people and the corresponding radar credibility are obtained from the detection millimeter-wave radar. For each detector group, the 6-degree-of-freedom detection values obtained by the detection millimeter-wave radar and the visual camera are weighted and averaged according to the credibility of the two to obtain the 6-degree-of-freedom result of the detector group. The detection results of the detector groups on different walls and the detection millimeter-wave radar on the ceiling are then combined, and the 6-degree-of-freedom results of each detection are summed and averaged. Finally, the 6-degree-of-freedom data results of the target people in the world coordinate system, including the position and posture angles, are obtained. The 6 degrees of freedom refer to the spatial position and posture of the human body determined by the 3D coordinates of the three key points of the target person's head and shoulders, including the 3D position (x, y, z) and the rotation angles (α, β, γ) of the spatial position.

2. The method for indoor human body spatial positioning and posture contour recognition according to claim 1 is characterized in that in the detector group, the visual camera obtains the spatial 6-DOF detection value of the key point of the target person {P xcamera , P ycamera , P zcamera , P αcamera , P βcamera , P γcamera }, corresponding to the visual credibility T camera , the detection millimeter wave radar obtains the target person's 6-degree-of-freedom detection value {P xradar , P yradar , P zradar , P αradar , P βradar , P γradar }, corresponding to radar credibility T radar , all the credibility values are assigned initial values, and the z-axis direction detection value P obtained in real time in this group of detectors is used. zcamera and P zradar As a benchmark, the real-time credibility of the detection millimeter-wave radar and visual camera is calculated according to the following formula: d' is the average distance from the detector group to the opposite wall plane; The 6-DOF result P of the detector group group as follows: in, The subscripts x, y, z, α, β, and γ of the degree of freedom value P correspond to the six degrees of freedom respectively. camera represents the value obtained by the visual camera, radar represents the value obtained by the millimeter-wave radar, and group represents the value of the detector group.

3. The method for indoor human body spatial positioning and posture contour recognition according to claim 2, characterized in that When the visual credibility of any detector T camera Or radar credibility T radar When the value is less than 0.5, the 6-DOF detection value detected by the detector does not participate in the 6-DOF result calculation of the detector group, that is, it is 0.

4. A method for indoor human body spatial positioning and posture contour recognition according to claim 2 or 3, characterized in that For a detector group, if T camera <T radar , then retain T radar and the 6-DOF detection value detected by its corresponding detector, otherwise T is retained camera And its corresponding 6-DOF detection value detected by the detector.

5. The method for indoor human body spatial positioning and posture contour recognition according to claim 1, characterized in that The detection results are converted to the world coordinate system. When the detection results of the detector groups on different walls and the detection results of the top surface detection millimeter wave radar are combined, the top surface detection millimeter wave radar only uses the position information to participate in the calculation. The target person's 6-degree-of-freedom data results P in the world coordinate system include position and posture angles. world The calculation formula is: n represents the number of detector groups, {P xgroup_i , P ygroup_i , P zgroup_i , P αgroup_i , P βgroup_i , P γgroup_i } represents the 6-DOF result of the i-th detector group, {P xtop_radar , P ytopradar , P ztop_radar } represents the three-dimensional position detection result of the top surface detection millimeter wave radar, and the final P world ={P xworld , P yworld , P zworld , P αworld , P βworld , P γworld }.

6. The method for indoor human body spatial positioning and posture contour recognition according to claim 1, wherein After setting up detector groups on different walls, a calibration is performed first. A marker cube marked with a QR code is placed in the center of the room floor. The detector groups are started and the detection results of each detector group on the marker cube are synchronized. The visual camera and millimeter-wave radar both output 6-degree-of-freedom information on the position and angle of the marker cube, forming a vector of [x, y, z, α, β, γ] to obtain the height from the ground to the top surface and the average distance from the detector group to the opposite wall. The above data will serve as the basic reference value for the subsequent calculation of the detector credibility and as the initialization data for the recognition system. The conversion matrix is obtained through calibration, which is used to convert the position and rotation information in different detector coordinate systems to the world coordinate system, where the position of the marker cube is the origin.

7. The method for indoor human body spatial positioning and posture contour recognition according to claim 1, characterized in that During the real-time detection process of the detector group, the original angle estimates of the three key points of the head and shoulders are corrected so that the angle estimates of the key points are synchronized with the angle values of the human body posture. Specifically: Based on the 3D coordinates of the three different lines of the head and two shoulders, the normal vector of the plane formed by the three points and the rotation angle of the normal vector corresponding to the three coordinate axes in the world coordinate system are calculated, thereby forming an estimated value of the human body posture. On this basis, the estimated value of the human body posture is used to correct the angles α, β, and γ of the three key points: ΔR jα =R jα -R humanα ΔR jβ =R jβ -R humanβ ΔR jγ =R jγ -R humanγ R′ jα =R jα +ΔR jα R′ jβ =R jβ +ΔR jβ R′ jγ =R jγ +ΔR jγ j is the number of the key point, j = 1, 2, 3, R jα 、R jβ 、R jγ is the angle value obtained for key point j, R humanα 、R humanβ 、R humanγ is the rotation angle obtained by the normal vectors of the three key points, R' jα , R' jβ , R' jγ It represents the corrected angle value of the jth key point. First, the difference between the estimated human posture angle and the angle calculated by the original key point is calculated. Then, this difference is used as the angle correction value for the next key point angle calculation, and the angle calculated by the key point is continuously corrected.

8. The method for indoor human body spatial positioning and posture contour recognition according to claim 1, wherein When all visual cameras fail to track the target person, that is, the values returned by all detectors are null, or the credibility of the values returned by the detectors is lower than the set threshold, the hotspot millimeter-wave radar is activated to continuously search for the human body position in the room at the set time interval until the new position of the target person is found.

Images