Fall detection method based on video sequence number
By acquiring key point data of the target object, utilizing Euclidean distance and angle analysis, and combining it with the k-means clustering algorithm, the accuracy and comprehensiveness issues of existing fall detection methods are solved, achieving more efficient fall detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN HONGDIAN TECH CORP
- Filing Date
- 2024-03-29
- Publication Date
- 2026-06-19
AI Technical Summary
Existing fall detection methods suffer from low accuracy and incomplete analysis, which can easily lead to misjudgments.
By acquiring data from multiple key points of the target body, the k-means clustering algorithm is used to divide the target body into predefined clusters. The Euclidean distance and angle between the key points and the cluster centers are calculated to determine whether the key points are outliers. The angle of the body parts is then combined to determine whether a fall has occurred.
It improves the accuracy and comprehensiveness of fall detection, reduces false alarms, lowers the misidentification rate of complex actions, and reduces deployment and maintenance costs.
Smart Images
Figure CN118279984B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer vision technology, and in particular to a fall detection method based on video sequence numbers. Background Technology
[0002] Falls are common in daily life and work, and can easily lead to fractures, concussions, soft tissue injuries, and other serious health risks. Therefore, it is crucial to detect and prevent falls from causing further harm.
[0003] Fall detection technology is a technique that uses sensors, machine learning, and data analytics to monitor people's fall behavior. It typically uses accelerometers, gyroscopes, and other sensors to detect sudden body movements or unusual postures to identify potential fall scenarios. These sensors can be embedded in smartphones, wearable devices, health monitoring devices, or environments such as home monitoring systems. When performing fall behavior detection, monitoring systems are prone to misjudgments when faced with more complex movements such as running, jumping, squatting, or bending over, because these movements cause rapid, dynamic changes in key body points that may resemble certain characteristics of a fall. This complexity increases the difficulty of the recognition algorithm, making it easier to confuse movement types during the identification process and increasing the likelihood of misjudgment.
[0004] In the process of realizing this invention, the inventors discovered at least the following problems in the prior art:
[0005] Existing fall detection methods suffer from low accuracy and incomplete analysis, which can easily lead to misjudgments. Summary of the Invention
[0006] The purpose of this invention is to provide a fall detection method based on video sequence numbers, thereby addressing the problems of low accuracy and incomplete analysis in existing fall detection methods, which easily lead to misjudgments. The various technical effects of the preferred solutions among the many technical solutions provided by this invention are detailed below.
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] This invention provides a fall detection method based on video sequence number, comprising:
[0009] S101. Obtain data information of multiple preset clusters and multiple key points of the target body in the current frame;
[0010] S102. In each key point, the target cluster of the key point is obtained based on the data information of the key point and the cluster center coordinates of each of the multiple preset clusters. Then, the target distance of the key point is obtained based on the key point and the cluster center of the target cluster of the key point. The target distance is the distance from the key point to the cluster center of the target cluster of the key point.
[0011] S103. If the target distance of the key point is greater than the distance threshold of the cluster center of the target cluster of the key point, then the key point is determined to be an anomaly.
[0012] S104. After performing steps S102 and S103 for each key point, the determination result of each key point is obtained;
[0013] S105. If the determination result of each key point is an anomaly, then the target body exhibits a falling behavior.
[0014] Preferably, step S102 includes:
[0015] Based on the data information of the key points and the cluster center coordinates of each of the multiple preset clusters, the Euclidean distance between the key points and the cluster center of each preset cluster is determined.
[0016] The minimum Euclidean distance between the key point and the cluster center of each preset cluster is determined as the target distance, and the formula for obtaining the target distance is as follows:
[0017]
[0018] Among them, D min denoted as the target distance of the key point, b as the coordinates of the key point, c as the coordinates of the cluster center, and k as the preset cluster number.
[0019] Preferably, after step S102, the method further includes:
[0020] S106. In each key point, if the cluster center of the target cluster of the key point does not belong to the actual cluster center of the key point, then the key point is determined to be an anomaly.
[0021] After performing step S106 or S103 for each key point, the determination result of each key point is obtained.
[0022] Preferably, the fall detection method further includes:
[0023] S107. After step S104, or if not all of the determined key points are abnormal points, obtain multiple included angles of the target body based on the coordinates of each key point, wherein the multiple included angles include the angle ∠ between the upper body of the target body and the horizontal axis. up The angle ∠ between the thigh and calf of the target body knee The angle ∠ between the thigh of the target body and the horizontal axis. Leg The angle ∠ between the lower leg of the target body and the horizontal axis. calf ;
[0024] S108. Based on each included angle, determine the state result of the body part corresponding to each included angle;
[0025] S109. In each angle, if the state of the body part corresponding to the angle is abnormal, and the key point of the body part corresponding to the angle is an abnormal key point, then the target body has fallen.
[0026] S110. After performing step S109 at each included angle, the detection result of the target body is obtained.
[0027] Preferably, the angle ∠ between the upper body of the target and the horizontal axis is obtained. up ,include:
[0028] Obtain the coordinates of multiple key points on the upper body of the target body, and determine the slope of the upper body of the target body;
[0029] Based on the slope of the upper body of the target, determine the angle ∠ between the upper body of the target and the horizontal axis. up ;
[0030]
[0031] ∠ up =arctan(m);
[0032] Where m is the slope of the upper body of the target, x is the x-coordinate of the key point, y is the y-coordinate of the key point, n is the number of key points, and i is any one of the n key points.
[0033] Preferably, the angle ∠ between the lower leg of the target body and the horizontal axis is obtained. calf ,include:
[0034] Acquire key points at both ends of the lower leg of the target body to determine the slope of the lower leg of the target body;
[0035] Based on the slope of the target body's lower leg, determine the angle ∠ between the target body's lower leg and the horizontal axis. calf ;
[0036]
[0037] ∠ calf =arctan(m1);
[0038] Where m1 is the slope of the lower leg of the target body, (x1, y1) are the coordinates of the key point on the first end of the lower leg of the target body, and (x2, y2) are the coordinates of the key point on the second end of the lower leg of the target body;
[0039] Obtain the angle ∠ between the thigh of the target body and the horizontal axis. Leg ,include:
[0040] Obtain key points at both ends of the target body's thigh to determine the slope of the target body's thigh;
[0041] Based on the slope of the target body's thigh, determine the angle ∠ between the target body's thigh and the horizontal axis. Leg ;
[0042]
[0043] ∠ Leg =arctan(m2);
[0044] Where m2 is the slope of the thigh of the target body, (x3, y3) are the coordinates of the key point on the first end of the thigh of the target body, and (x4, y4) are the coordinates of the key point on the second end of the thigh of the target body.
[0045] Preferably, the angle ∠ between the thigh and calf of the target body is obtained. knee ,include:
[0046] Based on the slope of the lower leg and the slope of the upper leg of the target body, determine the angle ∠ between the upper and lower legs of the target body. knee ;
[0047] ∠ knee = (m2-m1) / (1+m2m1).
[0048] Preferably, step S108 includes:
[0049] When ∠ knee If the angle is less than 90°, the state of the thigh and the lower leg is determined to be abnormal.
[0050] When ∠ up If the angle is less than 45°, then the state of the upper body is determined to be an abnormal state;
[0051] When ∠ LegIf the angle is less than 30°, the condition of the thigh is determined to be abnormal.
[0052] When ∠ calf If the angle is less than 60°, the condition of the lower leg is determined to be abnormal.
[0053] Preferably, step S109 includes:
[0054] When the included angle ∠ knee The corresponding states of the thigh and the calf are abnormal, and the included angle ∠ knee If both the corresponding knee and ankle key points are abnormal key points, then the target body is exhibiting a falling behavior.
[0055] Or when the included angle is ∠ up The corresponding upper body state is an abnormal state, and the included angle ∠ up If both the corresponding shoulder and hip key points are abnormal key points, then the target body is exhibiting a falling behavior.
[0056] Or when the included angle is ∠ Leg The corresponding thigh state result is an abnormal state, and the included angle ∠ Leg If both the corresponding key points of the hip and the key points of the knee are abnormal, then the target body is exhibiting a falling behavior.
[0057] Or when the included angle is ∠ Leg The corresponding thigh state result is an abnormal state and the included angle ∠ calf The corresponding lower leg state is abnormal, and the included angle ∠ Leg and the included angle ∠ calf If both the corresponding knee and ankle key points are abnormal key points, then the target body is exhibiting a falling behavior.
[0058] Preferably, after the target body falls, the method further includes:
[0059] When the target body is in a falling behavior, detect whether the duration of the falling behavior exceeds a preset time;
[0060] If so, the target body is determined to be in a fallen state.
[0061] Implementing one of the above-described technical solutions of the present invention has the following advantages or beneficial effects:
[0062] This invention acquires multiple key points of a target object. If each acquired key point is detected as an anomaly, the target object is considered to have fallen. By detecting each acquired key point, the accuracy and comprehensiveness of the fall detection method can be improved. Attached Figure Description
[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:
[0064] Figure 1 This is a first flowchart of an embodiment of the fall detection method based on video sequence number of the present invention;
[0065] Figure 2 This is a second flowchart of an embodiment of the fall detection method based on video sequence number of the present invention;
[0066] Figure 3 This is a schematic diagram of the key detection points in an embodiment of the fall detection method based on video sequence number of the present invention. Detailed Implementation
[0067] To make the objectives, technical solutions, and advantages of the present invention clearer, various exemplary embodiments described below will be referenced to the accompanying drawings, which form part of the exemplary embodiments, illustrating various exemplary embodiments that may be used to implement the present invention. Unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. It should be understood that they are merely examples of processes, methods, and apparatuses consistent with some aspects of the present invention disclosed as detailed in the appended claims, and other embodiments may be used, or structural and functional modifications may be made to the embodiments listed herein without departing from the scope and spirit of the present invention.
[0068] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," etc., indicate the orientation or positional relationship based on the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the referred element must have a specific orientation, or be constructed and operated in a specific orientation. The terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. The term "multiple" means two or more. The terms "connected" and "linked" should be interpreted broadly, for example, they can be fixed connections, detachable connections, integral connections, mechanical connections, electrical connections, communication connections, direct connections, indirect connections through an intermediate medium, and can be the internal connection of two elements or the interaction relationship between two elements. The term "and / or" includes any and all combinations of one or more of the related listed items. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0069] To illustrate the technical solution described in this invention, specific embodiments are described below, showing only the parts related to the embodiments of this invention.
[0070] Example 1:
[0071] like Figure 1 As shown, the present invention provides a fall detection method based on video sequence number, including:
[0072] S101. Obtain data information of multiple preset clusters and multiple key points of the target body in the current frame;
[0073] S102. In each key point, based on the data information of the key point and the cluster center coordinates of each of the multiple preset clusters, the target cluster of the key point is obtained. Then, based on the key point and the cluster center of the target cluster of the key point, the target distance of the key point is obtained. The target distance is the distance from the key point to the cluster center of the target cluster of the key point.
[0074] S103. If the target distance of a key point is greater than the distance threshold of the cluster center of the target cluster of the key point, then the key point is determined to be an outlier.
[0075] S104. After performing steps S102 and S103 for each key point, the determination result of each key point is obtained;
[0076] S105. If the determination result of each keypoint is an anomaly, then the target body exhibits a falling behavior. Specifically, by acquiring video frames of the target body, multiple keypoints of the target body are obtained. For each keypoint, it is assigned to the target cluster closest to the center of multiple preset clusters. When the target distance of a keypoint is detected to be greater than the distance threshold of the cluster center of the target cluster, the keypoint is an anomaly; otherwise, the keypoint is a normal point. After assigning and detecting each keypoint, if the determination result of each keypoint is an anomaly, then the target body exhibits a falling behavior.
[0077] This invention acquires multiple key points of a target object. If each acquired key point is detected as an anomaly, the target object is considered to have fallen. By detecting each acquired key point, the accuracy and comprehensiveness of the fall detection method can be improved.
[0078] At the same time, this application does not require a large amount of data to train the model, which reduces the complexity of deployment and maintenance and lowers costs.
[0079] As an optional implementation, step S101 includes:
[0080] The k-means clustering algorithm is used to divide the body parts of multiple key points on the target object into multiple preset clusters, and each key point is assigned to a corresponding cluster. For example, when acquiring key points at the shoulder, hip, knee, and ankle of the target object, the number of clusters is determined to be 4, namely the shoulder cluster, the hip cluster, the knee cluster, and the ankle cluster. The data information of multiple key points of the target object in the current frame is acquired. For each key point, the data information includes the coordinates of the key point and the displacement vector ΔP. The displacement vector ΔP is the displacement value of the key point in the current frame relative to the corresponding key point in the previous frame. The formula for obtaining the displacement vector ΔP is: ΔP = P(t) - P(t-1), where P(t) is the coordinate of the key point in the current frame, P(t-1) is the coordinate of the corresponding key point in the previous frame, and t is the frame number acquired.
[0081] More specifically, the velocity and acceleration of the keypoint in the current frame relative to the corresponding keypoint in the previous frame are determined based on the displacement vector ΔP. The formula for obtaining the velocity of the keypoint is: v = ΔP / Δt, where Δt is the time interval between the current frame and the previous frame, and v is the velocity of the keypoint. The formula for obtaining the acceleration of the keypoint is: a = Δv / Δt, where Δv = (v(t) - v(t-1)), which is the change in velocity, and a is the acceleration of the keypoint. The direction of motion of the keypoint is determined based on the displacement vector ΔP = (dx, dy). The formula for obtaining the direction of motion of the keypoint is: Where dx is the horizontal displacement of the key point, dy is the vertical displacement of the key point, and θ is the direction of movement of the key point.
[0082] Within each preset cluster, the cluster center is determined using the acceleration and direction of motion of each key point in the preset cluster through the k-means clustering algorithm.
[0083] By pre-acquiring multiple video frames without anomalies, the average value of all key points within each cluster is calculated, or the average value of all key points within each cluster is calculated every time a new video frame without anomalies is acquired, until the cluster center no longer changes or the preset number of iterations is reached, and the final average value is determined as the cluster center.
[0084] As an optional implementation, step S102 includes:
[0085] Based on the data information of the key points and the cluster center coordinates of each of the multiple preset clusters, the Euclidean distance between the key points and the cluster center of each preset cluster is determined.
[0086] The minimum Euclidean distance between the keypoint and the cluster center of each preset cluster is determined as the target distance. The formula for obtaining the target distance is as follows:
[0087]
[0088] Among them, D min Let 'b' be the target distance for the keypoint, 'c' be the coordinates of the keypoint, 'c' be the coordinates of the cluster center, and 'k' be the cluster number of the preset cluster. Specifically, when the preset clusters include shoulder clusters, hip clusters, knee clusters, and ankle clusters, the cluster center of the shoulder cluster is denoted as c1, the cluster center of the hip cluster as c2, the cluster center of the knee cluster as c3, and the cluster center of the ankle cluster as c4. Based on the coordinates of the keypoint and the coordinates of the shoulder cluster center, determine the Euclidean distance between the keypoint and the shoulder cluster center. Based on the coordinates of the keypoint and the coordinates of the hip cluster center, determine the Euclidean distance between the keypoint and the hip cluster center. Based on the coordinates of the keypoint and the coordinates of the knee cluster center, determine the Euclidean distance between the keypoint and the ankle cluster center. Based on the coordinates of the keypoint and the coordinates of the ankle cluster center, determine the Euclidean distance between the keypoint and the ankle cluster center. Select the minimum distance from the above Euclidean distances as the target distance. The formula for obtaining the target distance is as follows: Assign keypoints to the clusters whose cluster centers are at the target distance.
[0089] As an optional implementation, if the target distance of a keypoint is greater than the distance threshold of the cluster center of the target cluster of the keypoint, then the keypoint is determined to be an outlier. Specifically, each cluster has a cluster center distance threshold, which is the average Euclidean distance of all keypoints in the cluster from the cluster center. The formula for obtaining the cluster center distance threshold for each cluster is as follows: M is the number of keypoints within each cluster, r is any one of the M keypoints, and d r Let be the distance between the keypoint numbered r and the cluster center of the cluster. When the target distance D of the keypoint... min The distance threshold D of the cluster center of the target cluster greater than the key point avg At that time, the critical point was identified as an outlier.
[0090] As an optional implementation, after step S102, the method further includes:
[0091] S106. In each key point, if the cluster center of the target cluster of the key point does not belong to the actual cluster center of the key point, then the key point is determined to be an outlier.
[0092] After performing step S106 or S103 on each keypoint, the determination result for each keypoint is obtained. Specifically, the actual cluster center of a keypoint is the cluster center of the cluster in which the body part where the keypoint is located is divided. Since the specific body part of the target body is known when acquiring keypoints, the actual cluster center of the keypoint can be determined. When the cluster center of the target cluster of a keypoint does not belong to the actual cluster center of the keypoint, the keypoint is determined as an outlier. After performing step S106 or S103 on each acquired keypoint, the determination result of whether each keypoint is an outlier can be obtained.
[0093] As an optional implementation, the fall detection method further includes:
[0094] S107. After step S104, or if not all key points are identified as outliers, obtain multiple included angles of the target body based on the coordinates of each key point. These included angles include the angle ∠ between the upper body of the target body and the horizontal axis. up The angle ∠ between the thigh and calf of the target body knee The angle ∠ between the target's thigh and the horizontal axis Leg The angle ∠ between the target's lower leg and the horizontal axis calf ;
[0095] S108. Based on each included angle, determine the state result of the body part corresponding to each included angle;
[0096] S109. In each angle, if the state of the body part corresponding to the angle is abnormal, and the key point of the body part corresponding to the angle is an abnormal key point, then the target body has fallen behavior.
[0097] S110. After performing step S109 for each angle, the detection result of the target body is obtained. Specifically, after step S104, or if not all key points are determined to be abnormal points, based on the coordinates of multiple key points, the angle between the target body part (such as the upper body, thigh, calf, etc.) corresponding to the key point and the horizontal axis, and the angle between the target body parts corresponding to the key point and each other (such as the angle between the thigh and calf) are determined. The state result of the body part corresponding to each angle is detected to detect whether there is an abnormal state of the target body. Since human posture is relatively complex and there are actions similar to falling, such as running and jumping, step S110 is used to detect the abnormal state of the body part corresponding to each angle, and to detect the presence of abnormal key points of the key points of the body part corresponding to the angle, to ensure comprehensive detection of the target body, improve the accuracy of target body fall detection, reduce false judgments, and better protect the safety of the target body. Alternatively, after performing step S106 or S103 at each key point and obtaining the determination result of each key point, multiple included angles of the target body can be obtained based on the coordinates of each key point.
[0098] As an optional implementation, the angle ∠ between the upper body of the target and the horizontal axis is obtained. up ,include:
[0099] Obtain the coordinates of multiple key points on the upper body of the target object to determine the slope of the upper body of the target object;
[0100] Determine the angle ∠ between the upper body of the target and the horizontal axis based on the slope of the upper body. up ;
[0101]
[0102] ∠ up =arctan(m);
[0103] Where m is the slope of the upper body of the target, x is the x-coordinate of the keypoint, y is the y-coordinate of the keypoint, n is the number of keypoints, and i is any one of the n keypoints. Specifically, the preferred method for obtaining the coordinates of multiple keypoints of the upper body of the target includes the coordinates of the keypoints at the shoulders, neck, and hips. By obtaining the coordinates of multiple keypoints of the upper body of the target, the intercept of the upper body of the target is determined. The formula for obtaining the intercept of the upper body of the target is as follows: The prediction of the target's motion trajectory can be verified by using the intercept c of the target's upper body and the slope m of the target's upper body.
[0104] As an optional implementation, the angle ∠ between the target body's lower leg and the horizontal axis is obtained.calf ,include:
[0105] Obtain key points at both ends of the target's lower leg to determine the slope of the target's lower leg;
[0106] Determine the angle ∠ between the target's lower leg and the horizontal axis based on the slope of the target's lower leg. calf ;
[0107]
[0108] ∠ calf =arctan(m1);
[0109] Where m1 is the slope of the target's lower leg, (x1, y1) are the coordinates of the keypoints at the first end of the target's lower leg, and (x2, y2) are the coordinates of the keypoints at the second end of the target's lower leg. Specifically, the keypoints at the knee and ankle are obtained respectively. This is based on the angle ∠ between the target's lower leg and the horizontal axis. calf Determine the degree of inclination of the lower leg to facilitate the identification of any abnormalities in the body's posture through the lower leg's position.
[0110] As an optional implementation, the angle ∠ between the target body's thigh and the horizontal axis is obtained. Leg ,include:
[0111] Obtain key points at both ends of the target's thighs to determine the slope of the target's thighs;
[0112] Determine the angle ∠ between the target's thigh and the horizontal axis based on the slope of the target's thigh. Leg ;
[0113]
[0114] ∠ Leg =arctan(m2);
[0115] Where m2 is the slope of the target's thigh, (x3, y3) are the coordinates of the keypoint at the first end of the target's thigh, and (x4, y4) are the coordinates of the keypoint at the second end of the target's thigh. Specifically, the keypoints at the two ends of the target's thigh are obtained at the hip and knee, respectively. This is based on the angle ∠ between the target's thigh and the horizontal axis. Leg Determine the degree of inclination of the thighs to facilitate the identification of abnormal body posture through the posture of the thighs.
[0116] As an optional implementation, the angle ∠ between the thigh and calf of the target body is obtained. knee ,include:
[0117] Determine the angle ∠ between the target's thigh and lower leg based on the slopes of the target's lower leg and thigh. knee ;
[0118] ∠ knee = (m2-m1) / (1+m2m1). Specifically, the angle ∠ between the thigh and lower leg of the target body. knee The formula for obtaining ∠ is shown below: knee = (m2-m1) / (1+m2m1), where m1 is the slope of the target body's lower leg and m2 is the slope of the target body's upper leg.
[0119] As an optional implementation, step S108 includes: when ∠ knee When the angle is less than 90°, the condition of the thigh and calf is determined to be abnormal. When ∠ up When the angle is less than 45°, the upper body's condition is determined to be abnormal. When ∠ Leg When the angle is less than 30°, the thigh's condition is determined to be abnormal. When ∠ calf If the angle is less than 60°, the lower leg is determined to be in an abnormal state.
[0120] Specifically, based on each included angle and its corresponding included angle threshold, the state of the body part corresponding to each included angle is determined. The included angle threshold is adaptively set according to the actual situation. (The process involves) determining the included angle ∠. knee Is the angle between the thigh and lower leg of the target body less than the threshold value? If yes, the state of the thigh and lower leg is determined to be abnormal; otherwise, the state of the thigh and lower leg is determined to be normal. The preferred threshold value for the angle between the thigh and lower leg of the target body is 90°. Determine the angle ∠. up Is the angle between the upper body of the target object and the horizontal axis less than a threshold? If yes, the upper body's state is determined to be abnormal; otherwise, it is determined to be normal. The preferred threshold for the angle between the upper body of the target object and the horizontal axis is 45°. The angle ∠ is then used to determine the state. Leg Is the angle between the target body's thigh and the horizontal axis less than a threshold? If yes, the thigh's state is determined to be abnormal; otherwise, it is determined to be normal. The preferred threshold for the angle between the target body's thigh and the horizontal axis is 30°. The angle ∠ is then used to determine the result. calf If the angle between the target body's lower leg and the horizontal axis is less than a threshold, the lower leg's state is determined to be abnormal; otherwise, the lower leg's state is determined to be normal. The preferred threshold for the angle between the target body's lower leg and the horizontal axis is 60°.
[0121] As an optional implementation, step S109 includes:
[0122] When the included angle ∠ kneeThe corresponding thigh and calf conditions are abnormal, and the included angle ∠ knee If both the corresponding knee and ankle key points are abnormal, then the target body exhibits a falling behavior.
[0123] Or when the included angle is ∠ up The corresponding upper body state is abnormal, and the included angle ∠ up If both the corresponding shoulder and hip key points are abnormal key points, then the target body is exhibiting a falling behavior.
[0124] Or when the included angle is ∠ Leg The corresponding thigh's status result is abnormal, and the included angle ∠ Leg If both the hip and knee key points are abnormal key points, then the target body is exhibiting a falling behavior.
[0125] Or when the included angle is ∠ Leg The corresponding thigh status result is an abnormal state and the included angle ∠ calf The corresponding lower leg condition is abnormal, and the included angle ∠ Leg and the included angle ∠ calf If both the knee and ankle keypoints are abnormal, the target body exhibits a fall behavior. Specifically, within each angle, if the state of the body part corresponding to the angle is abnormal, and the keypoints of that body part are also abnormal, the target body exhibits a fall behavior. The keypoints of the body part corresponding to the angle can be adaptively determined based on the body part itself. In addition to the four methods mentioned above for determining if a target body is exhibiting a fall behavior, it can also be detected based on the actual situation, by ensuring that the state of the body part corresponding to the angle is abnormal, and that the keypoints on that body part, other than those mentioned above, are also abnormal.
[0126] As an optional implementation, after the target body exhibits a falling behavior, the method further includes: when the target body is falling, detecting whether the duration of the falling behavior exceeds a preset time; if so, determining that the target body is in a falling state. Specifically, by detecting the motion state of key points and the state of body parts, it is determined whether the target body is falling. When it is determined that the target body is falling, and the duration of the falling behavior exceeds the preset time, it is determined that the target body is in a falling state, and an alarm message is issued to provide an alarm notification. The preset time is an adaptively set time according to actual needs, preferably 1 second.
[0127] This invention acquires multiple key points of the target body, including but not limited to key points on the shoulders, hips, knees, and ankles, as well as key points on the elbows, wrists, nose, eyes, ears, and neck. Key point detection locations are as follows: Figure 3 As shown: 0 is the right shoulder, 1 is the left shoulder, 2 is the right elbow, 3 is the right wrist, 4 is the left elbow, 5 is the left wrist, 6 is the right hip, 7 is the right knee, 8 is the right ankle, 9 is the left hip, 10 is the left knee, 11 is the left ankle, 12 is the nose, 13 is the right eye, 14 is the right ear, 15 is the left eye, 16 is the left ear, and 17 is the neck.
[0128] The embodiment is merely a specific example and does not indicate that this is the only way to implement the present invention.
[0129] The above description is merely a preferred embodiment of the present invention. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the present invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.
Claims
1. A fall detection method based on video sequence number, characterized in that, include: S101. Obtain data information of multiple preset clusters and multiple key points of the target body in the current frame; S102. In each key point, the target cluster of the key point is obtained based on the data information of the key point and the cluster center coordinates of each of the multiple preset clusters. Then, the target distance of the key point is obtained based on the key point and the cluster center of the target cluster of the key point. The target distance is the distance from the key point to the cluster center of the target cluster of the key point. S103. If the target distance of the key point is greater than the distance threshold of the cluster center of the target cluster of the key point, then the key point is determined to be an anomaly. S104. After performing steps S102 and S103 for each key point, the determination result of each key point is obtained; S105. If the determination result of each key point is an anomaly, then the target body exhibits a falling behavior. S107. After step S104, or if not all of the determined key points are anomalies, obtain multiple included angles of the target body based on the coordinates of each key point, wherein the multiple included angles include the angle ∠ between the upper body of the target body and the horizontal axis. up The angle ∠ between the thigh and calf of the target body knee The angle ∠ between the thigh of the target body and the horizontal axis. Leg The angle ∠ between the lower leg of the target body and the horizontal axis. calf ; S108. Based on each included angle, determine the state result of the body part corresponding to each included angle; S109. In each angle, if the state of the body part corresponding to the angle is abnormal, and the key point of the body part corresponding to the angle is an abnormal key point, then the target body has fallen. S110. After performing step S109 at each included angle, the detection result of the target body is obtained.
2. The fall detection method based on video sequence number according to claim 1, characterized in that, Step S102 includes: Based on the data information of the key points and the cluster center coordinates of each of the multiple preset clusters, the Euclidean distance between the key points and the cluster center of each preset cluster is determined. The minimum Euclidean distance between the key point and the cluster center of each preset cluster is determined as the target distance, and the formula for obtaining the target distance is as follows: ； Where Dmin is the target distance of the key point, b is the coordinate of the key point, c is the coordinate of the cluster center, and k is the number of the preset cluster.
3. The fall detection method based on video sequence number according to claim 1, characterized in that, Following step S102, the method further includes: S106. In each key point, if the cluster center of the target cluster of the key point does not belong to the actual cluster center of the key point, then the key point is determined to be an anomaly. After performing step S106 or S103 for each key point, the determination result of each key point is obtained.
4. The fall detection method based on video sequence number according to claim 1, characterized in that, obtaining an angle ∠ of an upper body of the target object with a horizontal axis up comprising: Obtain the coordinates of multiple key points on the upper body of the target body, and determine the slope of the upper body of the target body; determining an angle ∠ of the upper body of the target object with a horizontal axis according to the slope of the upper body of the target object up ; ； ∠ up = arctan(m); Where m is the slope of the upper body of the target, x is the x-coordinate of the key point, y is the y-coordinate of the key point, n is the number of key points, and i is any one of the n key points.
5. The fall detection method based on video sequence number according to claim 1, characterized in that, obtaining an angle ∠ of a lower leg of the target body with a horizontal axis calf comprising: Acquire key points at both ends of the lower leg of the target body to determine the slope of the lower leg of the target body; determining an angle ∠ of the shank of the target object with a horizontal axis based on a shank slope of the target object calf ; ； ∠ calf = arctan(m1); Where m1 is the slope of the lower leg of the target body, (x1, y1) are the coordinates of the key point on the first end of the lower leg of the target body, and (x2, y2) are the coordinates of the key point on the second end of the lower leg of the target body; obtaining an angle ∠ between a thigh of the target body and a horizontal axis Leg comprising: Obtain key points at both ends of the target body's thigh to determine the slope of the target body's thigh; determining an angle ∠ of the thigh of the target with a horizontal axis according to a slope of the thigh of the target Leg ; ； ∠ Leg =arctan(m2); Where m2 is the slope of the thigh of the target body, (x3, y3) are the coordinates of the key point on the first end of the thigh of the target body, and (x4, y4) are the coordinates of the key point on the second end of the thigh of the target body.
6. The fall detection method based on video sequence number according to claim 5, characterized in that, Obtain the angle ∠ between the thigh and calf of the target body. knee ,include: Based on the slope of the lower leg and the slope of the upper leg of the target body, determine the angle ∠ between the upper and lower legs of the target body. knee ; ∠ knee =(m2-m1) / (1+m2m1) 7. The fall detection method based on video sequence number according to claim 1, characterized in that, Step S108 includes: When ∠ knee If the angle is less than 90°, the state of the thigh and the lower leg is determined to be abnormal. When ∠ up If the angle is less than 45°, then the state of the upper body is determined to be an abnormal state; When ∠ Leg If the angle is less than 30°, the condition of the thigh is determined to be abnormal. When ∠ calf If the angle is less than 60°, the condition of the lower leg is determined to be abnormal.
8. The fall detection method based on video sequence number according to claim 1, characterized in that, Step S109 includes: When the included angle ∠ knee The corresponding states of the thigh and the calf are abnormal, and the included angle ∠ knee If both the corresponding knee and ankle key points are abnormal key points, then the target body is exhibiting a falling behavior. Or when the included angle is ∠ up The corresponding upper body state is an abnormal state, and the included angle ∠ up If both the corresponding shoulder and hip key points are abnormal key points, then the target body is exhibiting a falling behavior. Or when the included angle is ∠ Leg The corresponding thigh state result is an abnormal state, and the included angle ∠ Leg If both the corresponding key points of the hip and the key points of the knee are abnormal, then the target body is exhibiting a falling behavior. Or when the included angle is ∠ Leg The corresponding thigh state result is an abnormal state and the included angle ∠ calf The corresponding lower leg state is abnormal, and the included angle ∠ Leg and the included angle ∠ calf If both the corresponding knee and ankle key points are abnormal key points, then the target body is exhibiting a falling behavior.
9. The fall detection method based on video sequence number according to claim 1, characterized in that, After the target body falls, the following is also included: When the target body is in a falling behavior, detect whether the duration of the falling behavior exceeds a preset time; If so, the target body is determined to be in a fallen state.