Non-human primate fine motion and posture tracking automation method

Abstract

The invention provides a non-human primate fine motion and posture tracking automation method, which is implemented by taking a monkey as a representative, and comprises the following steps: defining 10 motion and 3 posture labels of the monkey, defining 15 key points of the monkey body, constructing a monkey motion identification data set and a monkey key point data set, and establishing a monkey motion and posture identification analysis tool box. The monkeys are subjected to action recognition and skeleton point recognition, and the three postures of platform, sitting and curling up of the monkeys, the engraving behavior and the depression behavior are automatically recognized. According to the invention, automatic identification of fine actions and postures of the non-human primates can be effectively realized, and particularly, effective automatic identification of engraving behaviors, depression behaviors and the like of the non-human primates can be realized, so that the blank of such researches in the field is filled; the method is of great significance to clinical research of difficult miscellaneous diseases such as depression and Rett syndrome.

Description

Technical Field

[0001] The present invention relates to the technical field of neuroscience experimental research, and in particular to an automated method for tracking fine movements and postures of non-human primates. Background Art

[0002] Motion recognition based on non-human primate (NHP) videos is becoming an important tool for disease detection and prognosis in neuroscience and clinical research. Traditionally, animal motion recognition in NHP research relies primarily on manual labor, which is labor-intensive and inefficient. In recent years, research has been conducted using network models to recognize and analyze human motion. However, research on automatically recognizing larger movements in NHPs, such as walking, running, and jumping, is still in its infancy. Existing research models have difficulty automatically recognizing very subtle movements and fine gestures in NHP videos, such as head-down (a key phenotypic characteristic of depressive behavior) and stereotypic behaviors (RTT, the primary phenotypic characteristic of Rett syndrome). However, these fine movements and gestures are crucial for motion recognition and analysis in NHP videos and are of great significance for clinical research on difficult-to-diagnose diseases, including depression and Rett syndrome. Summary of the Invention

[0003] The purpose of the present invention is to provide an automated method for tracking fine movements and postures of non-human primates in order to solve the problems existing in the prior art.

[0004] The technical solution of the present invention is: an automated method for tracking fine movements and postures of non-human primates, wherein the method is implemented on monkeys as a representative non-human primate, and comprises the following steps:

[0005] The first step is to define 10 actions and 3 posture labels of the monkey;

[0006] The second step is to define 15 key points of the monkey's body;

[0007] The third step is to build a monkey action posture recognition dataset and a monkey key point dataset;

[0008] The fourth step is to build a monkey action posture recognition and analysis toolbox;

[0009] The fifth step is to perform action recognition on the monkey;

[0010] Step 6: Identify the monkey's skeleton points;

[0011] The seventh step is to automatically identify the monkey's standing, sitting and curled-up postures, as well as stereotyped and depressive behaviors.

[0012] A further solution is: In the first step above, based on the monkey's action behavior in the experimental cage, the monkey's 10 action labels are specifically defined as:

[0013] Climbing: Using the legs or arms, slowly move from the bottom of the cage to the side wall, or from the side wall to the top of the cage;

[0014] Hanging: hanging on a lever set on the top of the experimental cage, or moving from one lever to another lever set on the top of the cage;

[0015] Turning: bending at the waist or twisting with arms and legs from a standing position;

[0016] Walking: Stand on the bottom of the cage or on the horizontal bars in the middle of the cage and walk;

[0017] Shake: Stand with two feet in the cage and shake the body;

[0018] Jumping: Use the legs to quickly jump from the bottom of the cage into the air or onto the side wall of the experimental cage;

[0019] Move downward: quickly jump from the side wall of the experimental cage to the bottom of the cage;

[0020] Lying down: from standing on all fours to lying prone;

[0021] Sit: Change from any other posture to sitting or squatting;

[0022] Stand Up: Move from a sitting or squatting position to a standing position;

[0023] The three monkey posture labels are specifically defined as:

[0024] Platform: Stand on the middle crossbar of the experimental cage and look out;

[0025] Sitting platform: sitting on the middle horizontal bar of the experimental cage;

[0026] Huddle: Lie on the ground and curl up.

[0027] A further solution is: in the second step above, define the 15 key points of the monkey's body as: right ankle, right knee, left knee, left ankle, hip joint, tail, chin, top of the head, right wrist, right elbow, right shoulder, left shoulder, left elbow, left wrist, and neck.

[0028] A further solution is: the specific method of the third step is:

[0029] Use Python scripts to manually annotate the original collected videos and build an action recognition dataset that includes the 10 monkey actions defined in the first step;

[0030] The Python script tool is used to manually mark the 15 key points of the monkey body defined in the second step on the original collected images to obtain the monkey key point dataset.

[0031] A further solution is: the monkey action posture recognition and analysis toolbox established in the fourth step above includes the OpenCV tool and Dense_Flow tool for converting video clips into RGB images and optical flow images, a two-stream network based on time shift and piece group attention with a residual neural network as the backbone network for monkey action recognition, an HRNet network for obtaining key points on the monkey's body, a MaskTrack R-CNN model for tracking the monkey's position, and a Grad-Cam++ method for constructing monkey heat maps.

[0032] A further solution is: the specific method of the fifth step is:

[0033] The action recognition dataset constructed in the third step is used to train the two-stream network of the monkey action posture recognition and analysis toolbox established in the fourth step. The trained two-stream network is used to perform action recognition on unknown video data. Specifically, video clips are randomly sampled from the unknown original video, and the OpenCV tool and the Dense_Flow tool are used to extract RGB images and optical flow images for each sampled video clip respectively; the extracted RGB images and optical flow images are used as the input of the two-stream network, and the two-stream network analyzes and outputs the monkey action results corresponding to the 10 defined monkey action labels in the video data.

[0034] A further solution is: the specific method of the sixth step is:

[0035] The monkey key point dataset constructed in the third step is used to train the HRNet network of the monkey motion posture recognition and analysis toolbox established in the fourth step, and the MSE loss function is used to compare targets and calculate losses; the trained HRNet network is used to recognize skeleton points in unknown video data. Specifically, the unknown monkey video is input into the MaskTrack R-CNN model of the monkey recognition and analysis toolbox to track the monkey's position, and then a rectangular box representing the position of each monkey is intercepted. The rectangular box with the monkey is then used as input and input into the trained HRNet network model to obtain a heat map of 15 key points on the monkey's body. The key point of the monkey's neck is determined by taking the center of the left and right shoulders. Finally, the heat map is converted into the spatial coordinates of the x-axis and y-axis of the 15 key points.

[0036] A further solution is: the specific method of the seventh step includes the following steps:

[0037] ① Identify the three postures of the monkey: From the x-axis and y-axis spatial coordinates of the 15 key points established in step 6, obtain the y-axis coordinates of the monkey's right ankle, left ankle, hip joint, and top of the head at the same time point through the HRNet network. Select the maximum value ymax and the minimum value ymin of the four y-axis coordinates, and calculate the monkey's height information H using the formula H = ymax-ymin. If H is not less than 2 / 3 of the monkey's height, it is judged as a standing posture; if H is greater than 1 / 3 of the monkey's height and less than 2 / 3 of the monkey's height, it is judged as a sitting posture; if the value of H is less than 1 / 3 of the monkey's height, it is judged as a curled-up posture;

[0038] ②Identify the monkey’s stereotyped behavior:

[0039] Using the x-axis and y-axis spatial coordinates of the 15 key points determined in the sixth step, the coordinates of the monkey center point Ccenter(x,y) are calculated using formula (1):

[0040]

[0041] Where N is the number of key points, N=15; i ranges from 1 to 15;

[0042] Formula (2) is used to judge the stereotyped behavior of monkeys:

[0043]

[0044] In the formula, the motion vector V is defined as V = C center (t+1)-C center (t) represents the vector change of the monkey's center point in adjacent moments. The vector V in the x and y dimensions in a period of time is summed up and compared with the judgment threshold T. If the sum is less than the judgment threshold T, the monkey's behavior during this period is judged to be stereotyped behavior. The judgment threshold T is 1 / 10 of the input image size.

[0045] ③Identifying depressive behavior in monkeys:

[0046] Among the x-axis and y-axis space coordinates of the 15 key points determined in step 6, take the y-axis coordinate of the bone point corresponding to the monkey's chin chin The y-axis coordinate of the bone point corresponding to the neck neck , calculate the h value by formula (3):

[0047] h=y chin -y neck (3)

[0048] If h<0, the monkey is judged to be in a head-down posture, and is accordingly determined to be in depressive behavior.

[0049] The present invention has positive effects: the present invention innovatively establishes a non-human primate recognition and analysis toolbox, defines 13 types of non-human primate action and posture labels, and defines key points on the monkey's body. According to the defined actions and key point definitions, the collected original videos are manually annotated using a Python annotation script tool to construct an action recognition dataset and a key point dataset; then a two-stream network (TS-SA Net) based on time shift (TS) and split attention (SA) is used to perform action recognition on non-human primates; HRNet is used to perform skeleton recognition on non-human primates at the same time; based on the results of action recognition and skeleton recognition, a fine action modeling model is used to perform fine action and posture recognition. The present invention can effectively realize the effective automatic recognition and analysis of non-human primates, including head-down behavior, stereotyped behavior, and depressive behavior, filling the gap in such research in this field and having very important significance for the clinical research of difficult and complicated diseases such as depression and Rett syndrome. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] (Example 1)

[0053] The automated method for tracking fine movements and postures of non-human primates in this embodiment first defines 13 types of movements and posture labels of non-human primates and defines key points on the monkey's body. According to the defined movements and key point definitions, the collected original videos are manually annotated using a Python annotation script tool to construct a movement recognition dataset (referred to as MiL) and a key point dataset (referred to as MiL2D); then a non-human primate recognition and analysis toolbox (referred to as MonKit) is established, which includes tracking and recognition, movement recognition, posture recognition, key point recognition, and fine movement recognition functions for non-human primates. First, based on the movement recognition dataset (referred to as MiL), a two-stream network (TS-SA Net) based on time shift (TS) and split attention (SA) is used to perform movement recognition on non-human primates. At the same time, the MaskTrack R-CNN model is used to track and recognize the monkey, and then HRNet is used to perform skeleton recognition on the non-human primate. Based on the results of motion recognition and skeleton recognition, a fine motion modeling model is used for fine motion and posture recognition, which can effectively realize the automatic recognition and analysis of fine movements of non-human primates, including head-lowering behavior, stereotyped behavior, etc.; finally, manual detection is used to compare with the automatic detection results to verify the effectiveness of the automatic recognition results, thus innovatively providing an effective automated method for fine motion and posture tracking of non-human primates, filling the gap in such research in this field.

[0054] See also Figure 1 The automated method for tracking fine movements and postures of non-human primates of this embodiment is implemented using monkeys, which are the most commonly used non-human primates in this field, as representatives, and specifically includes the following steps:

[0055] The first step is to define 10 actions and 3 posture labels for the monkey:

[0056] According to the monkey's action behavior in the experimental cage, 10 action labels and 3 posture labels are defined for the monkey.

[0057] The specific definitions of the 10 action tags are as follows:

[0058] Climbing: Using the legs or arms, slowly move from the bottom of the cage to the side wall, or from the side wall to the top of the cage;

[0059] Hanging: hanging on a lever set on the top of the experimental cage, or moving from one lever to another lever set on the top of the cage;

[0060] Turning: bending at the waist or twisting with arms and legs from a standing position;

[0061] Walking: Stand on the bottom of the cage or on the horizontal bars in the middle of the cage and walk;

[0062] Shake: Stand with two feet in the cage and shake the body;

[0063] Jumping: Use the legs to quickly jump from the bottom of the cage into the air or onto the side wall of the experimental cage;

[0064] Move downward: quickly jump from the side wall of the experimental cage to the bottom of the cage;

[0065] Lying down: from standing on all fours to lying prone;

[0066] Sit: Change from any other posture to sitting or squatting;

[0067] Stand Up: Move from a sitting or squatting position to a standing position.

[0068] The three posture labels are specifically defined as follows:

[0069] Platform: Stand on the middle crossbar of the experimental cage and look out;

[0070] Sitting platform: sitting on the middle horizontal bar of the experimental cage;

[0071] Huddle: Lie on the ground and curl up.

[0072] The second step is to define 15 key points of the monkey's body:

[0073] Define 15 parts of the monkey's body as key points of the monkey's body, including: right ankle, right knee, left knee, left ankle, hip joint, tail, chin, top of the head, right wrist, right elbow, right shoulder, left shoulder, left elbow, left wrist, and neck.

[0074] The third step is to build the monkey action recognition dataset and the monkey key point dataset:

[0075] The existing Python script is used to manually annotate the original collected videos to construct a motion recognition dataset (MiL) including 10 kinds of monkey actions.

[0076] The original collected images were manually labeled with 15 key points using a Python script tool to obtain the Monkey Key Point Dataset (MiL2D).

[0077] The fourth step is to build a monkey action posture recognition and analysis toolbox:

[0078] The recognition analysis toolbox includes the existing OpenCV tools and Dense_Flow tools for converting video clips into RGB images and optical flow images, the existing two-stream network (TS-SA Net network) based on temporal shift (TS) and split attention (SA) with the residual neural network (ResNet) as the backbone network for monkey action recognition, the existing HRNet network for obtaining key points on the monkey's body, the existing MaskTrack R-CNN model for tracking the monkey's position, and the existing Grad-Cam++ method (Chattopadhay et al., 2018, a feature visualization method for CNN model prediction) for constructing monkey heat maps.

[0079] Step 5: Recognize the monkey’s actions:

[0080] Based on the action recognition dataset (MiL), the TS-SA network is trained and the trained network is used to perform action recognition on unknown video data. The specific process is as follows:

[0081] Video clips are randomly sampled from unknown original videos. OpenCV tools and Dense_Flow tools are used to extract RGB images and optical flow images for each sampled video clip. The extracted RGB images and optical flow images are used as inputs of the TS-SA network, and 10 monkey action results are obtained through the TS-SA network.

[0082] Step 6: Identify the monkey's skeleton:

[0083] Based on the Monkey Keypoint Dataset (MiL2D), the HRNet network is trained and the MSE loss function is used to compare targets and calculate losses. The trained network is used to identify skeleton points in unknown video data. The specific process is as follows:

[0084] The MaskTrack R-CNN model is input into the unknown monkey video to track the monkey's position. Then, the rectangular information representing the position of each monkey is intercepted. The rectangular box with the monkey is used as input and input into the HRNet network model to obtain a heat map of 15 key points (also known as bone points) on the monkey's body (the neck is determined by taking the center of the left and right shoulders). Finally, the heat map is converted into the x and y spatial coordinates of the 15 key points.

[0085] The seventh step is to automatically identify the monkey's standing, sitting, and curled-up postures, as well as stereotyped and depressive behaviors, including the following steps:

[0086] ① Identify the three postures of monkeys:

[0087] From the x and y spatial coordinates of the 15 key points established in step 6, the HRNet network was used to obtain the y-axis coordinates of the monkey's right ankle, left ankle, hip joint, and top of the head at the same time. The maximum value ymax and the minimum value ymin of the four y-axis coordinates were selected, and the monkey's height information H was calculated using H = ymax - ymin. The monkey's standing, sitting, and crouching postures were then determined accordingly. Specifically, if H is at least 2 / 3 of the monkey's height, the posture is determined to be standing; if H is greater than 1 / 3 and less than 2 / 3 of the monkey's height, the posture is determined to be sitting; and if H is less than 1 / 3 of the monkey's height, the posture is determined to be crouching. Due to the extremely small amplitude of these three postures, they are difficult to identify using existing techniques.

[0088] ②Identify the monkey’s stereotyped behavior:

[0089] The stereotyped behavior of monkeys mainly refers to the repetitive and purposeless body movements of monkeys at a fixed frequency, including turning over, circling, pacing and shaking the cage, etc. This is a characteristic of autism and also represents the anxiety or hyperactivity state of monkeys, which is often observed in RTT patients.

[0090] In this embodiment, the x and y spatial coordinates of the 15 key points determined in step 6 are used to calculate the coordinates of the monkey's center point using formula (1):

[0091]

[0092] Wherein, N is the number of key points, in this embodiment, N=15; i ranges from 1 to 15;

[0093] Stereotyped behavior is characterized by repetitive movements, similar to circular motion, which means that the vector direction of a specific action should be equal to zero, indicating that the animal moves in a reciprocating motion pattern. Therefore, in this embodiment, formula (2) is used to determine the stereotyped behavior of the monkey:

[0094]

[0095] In the formula, the motion vector V is defined as V = C center (t+1)-C center (t) represents the vector change of the monkey's center point between adjacent moments. The sum of the x- and y-dimensional vectors V over a period of time is calculated and compared with a specific threshold T. If the sum is less than the threshold T, the monkey's behavior during that period is considered stereotyped. The value of T can be modified appropriately based on the specific situation; lower values are more stringent. In this embodiment, T is set to 1 / 10 of the input image size.

[0096] ③Identifying depressive behavior in monkeys:

[0097] In addition to the curled-up posture defined in the posture tag above, which is considered a depressive behavior, the fetal position in which the monkey's head is placed at or below the shoulders when awake is also considered a depressive behavior (Hryniewicka Jaworska, Foden et al., 2016). In this embodiment, among the x and y spatial coordinates of the 15 key points determined in step 6, the y-axis coordinate of the bone point corresponding to the monkey's chin is taken as y. chin The y-axis coordinate of the bone point corresponding to the neck neck , calculate the h value by formula (3):

[0098] h=y chin -y neck (3)

[0099] If the calculation result is h<0, the monkey is judged to be in a head-down posture, and thus its behavior is determined to be depressive.

[0100] To verify the accuracy and effectiveness of the automated method for tracking fine movements and postures of non-human primates of this embodiment, the original manual analysis was compared with its recognition analysis results. The experiment compared the differences between 11 movement categories of monkeys and found that the recognition error of this method for all monkey movements was always less than 5%, and the maximum detection error of monkey postures averaged 4.9%. At the same time, the results of manual recognition analysis of stereotyped behaviors and depressive behaviors of monkeys were basically similar to those of automatic recognition analysis of this method, thereby verifying the accuracy and effectiveness of the method of this embodiment.

[0101] The above embodiments are illustrations of specific implementation methods of the present invention, rather than limitations of the present invention. Technicians in the relevant technical fields can make various changes and modifications to obtain corresponding equivalent technical solutions without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions should be included in the patent protection scope of the present invention.

Claims

1. An automated method for tracking fine movements and postures of non-human primates, wherein the non-human primates are represented by monkeys, and wherein: The following steps are involved: The first step is to define 10 actions and 3 posture labels of the monkey; The second step is to define 15 key points of the monkey's body; The third step is to build a monkey action recognition dataset and a monkey key point dataset; The fourth step is to build a monkey action posture recognition and analysis toolbox; The fifth step is to perform action recognition on the monkey; Step 6: Identify the monkey's skeleton points; The seventh step is to automatically identify the monkey's standing, sitting and curled-up postures, as well as stereotyped and depressive behaviors.

2. The method for automating fine movement and posture tracking of non-human primates according to claim 1, wherein: In the first step, based on the monkeys' motor behaviors in the experimental cage, 10 action labels of monkeys were specifically defined as follows: climbing: slowly climbing from the bottom of the cage to the side wall, or moving from the side wall to the top of the cage by legs or arms; Hanging: hanging on a lever set on the top of the experimental cage, or moving from one lever to another lever set on the top of the cage; Turning: bending at the waist or twisting with arms and legs from a standing position; Walking: Stand on the bottom of the cage or on the horizontal bars in the middle of the cage and walk; Shake: Stand with two feet in the cage and shake the body; Jumping: Use the legs to quickly jump from the bottom of the cage into the air or onto the side wall of the experimental cage; Move downward: quickly jump from the side wall of the experimental cage to the bottom of the cage; Lying down: from standing on all fours to lying prone; Sit: Change from any other posture to sitting or squatting; Stand Up: Move from a sitting or squatting position to a standing position; The three monkey posture labels are specifically defined as: Platform: Stand on the middle crossbar of the experimental cage and look out; Sitting platform: sitting on the middle horizontal bar of the experimental cage; Huddle: Lie on the ground and curl up.

3. The method for automating fine movement and posture tracking of non-human primates according to claim 1, wherein: In the second step, 15 key points of the monkey's body are defined as: right ankle, right knee, left knee, left ankle, hip joint, tail, chin, top of the head, right wrist, right elbow, right shoulder, left shoulder, left elbow, left wrist, and neck.

4. The method for automating fine movement and posture tracking of non-human primates according to claim 1, wherein: The specific method of the third step is: Use Python scripts to manually annotate the original collected videos and build an action recognition dataset that includes the 10 monkey actions defined in the first step; The Python script tool is used to manually mark the 15 key points of the monkey body defined in the second step on the original collected images to obtain the monkey key point dataset.

5. The automated method for tracking fine movements and postures of non-human primates according to claim 1, wherein: The monkey action posture recognition and analysis toolbox established in the fourth step includes the OpenCV tool and the Dense_Flow tool for converting video clips into RGB images and optical flow images, a two-stream network based on time shift and slice group attention with a residual neural network as the backbone network for monkey action recognition, an HRNet network for obtaining key points on the monkey's body, a MaskTrack R-CNN model for tracking the monkey's position, and a Grad-Cam++ method for constructing a monkey heat map.

6. The method for automating fine movement and posture tracking of non-human primates according to claim 5, wherein: The specific method of the fifth step is: The action recognition dataset constructed in the third step is used to train the two-stream network of the monkey action posture recognition and analysis toolbox established in the fourth step. The trained two-stream network is used to perform action recognition on unknown video data. Specifically, video clips are randomly sampled from the unknown original video, and the OpenCV tool and the Dense_Flow tool are used to extract RGB images and optical flow images for each sampled video clip respectively; the extracted RGB images and optical flow images are used as the input of the two-stream network, and the two-stream network analyzes and outputs the monkey action results corresponding to the 10 defined monkey action labels in the video data.

7. The automated method for tracking fine movements and postures of non-human primates according to claim 5, wherein: The specific method of the sixth step is: The monkey key point dataset constructed in the third step is used to train the HRNet network of the Monkey Action Posture Recognition and Analysis Toolbox established in the fourth step, and the MSE loss function is used to compare targets and calculate losses; the trained HRNet network is used to recognize skeleton points in unknown video data. Specifically, the unknown monkey video is input into the MaskTrack R-CNN model of the Monkey Action Posture Recognition and Analysis Toolbox to track the monkey's position, and then a rectangular box representing the position of each monkey is intercepted. The rectangular box with the monkey is then used as input and input into the trained HRNet network model to obtain a heat map of 15 key points on the monkey's body. The key point of the monkey's neck is determined by taking the center of the left and right shoulders. Finally, the heat map is converted into the spatial coordinates of the x-axis and y-axis of the 15 key points.

8. The automated method for tracking fine movements and postures of non-human primates according to claim 7, wherein: The specific method of the seventh step includes the following steps: ① Identify the three postures of the monkey: From the x-axis and y-axis spatial coordinates of the 15 key points established in step 6, obtain the y-axis coordinates of the monkey's right ankle, left ankle, hip joint, and top of the head at the same time point through the HRNet network. Select the maximum value ymax and the minimum value ymin of the four y-axis coordinates, and calculate the monkey's height information H using the formula H = ymax-ymin. If H is not less than 2 / 3 of the monkey's height, it is judged as a standing posture; if H is greater than 1 / 3 of the monkey's height and less than 2 / 3 of the monkey's height, it is judged as a sitting posture; if the value of H is less than 1 / 3 of the monkey's height, it is judged as a curled-up posture; ②Identify the monkey’s stereotyped behavior: Using the x-axis and y-axis spatial coordinates of the 15 key points determined in the sixth step, the coordinates of the monkey center point Ccenter(x,y) are calculated using formula (1): Where N is the number of key points, N=15; i ranges from 1 to 15; Formula (2) is used to judge the stereotyped behavior of monkeys: In the formula, the motion vector V is defined as V=C center (t+1)-C center (t) represents the vector change of the monkey's center point in adjacent moments. The vector V in the x and y dimensions in a period of time is summed up and compared with the judgment threshold T. If the sum is less than the judgment threshold T, the monkey's behavior during this period is judged to be stereotyped behavior. The judgment threshold T is 1 / 10 of the input image size. ③Identifying depressive behavior in monkeys: Among the x-axis and y-axis space coordinates of the 15 key points determined in step 6, take the y-axis coordinate of the bone point corresponding to the monkey's chin chin The y-axis coordinate of the bone point corresponding to the neck neck , calculate the h value by formula (3): h=y chin -y neck (3) If h<0, the monkey is judged to be in a head-down posture, and is accordingly determined to be in depressive behavior.

Images