Skip to content
TheChatArchiveThe Chat Archive
Qwenqwen3-235b-a22b-no-thinking· Alibaba (Qwen)Hugging Face LMArena Human Preference 140K

import cv2 import mediapipe as mp import numpy as np import time import csv 初始化MediaPipe…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with gpt-4.1-mini-2025-04-14’s answer.
  1. User

    import cv2 import mediapipe as mp import numpy as np import time import csv

    初始化MediaPipe Pose模块

    mp_pose = mp.solutions.pose mp_drawing = mp.solutions.drawing_utils mp_drawing_styles = mp.solutions.drawing_styles

    存储角度数据

    left_shoulder_angles = [] # 左肩关节角度 right_shoulder_angles = [] # 右肩关节角度 left_elbow_angles = [] # 左肘关节角度 right_elbow_angles = [] # 右肘关节角度 left_hip_angles = [] # 左髋关节角度 right_hip_angles = [] # 右髋关节角度 left_knee_angles = [] # 左膝关节角度 right_knee_angles = [] # 右膝关节角度 left_ankle_angles = [] # 左踝关节角度 right_ankle_angles = [] # 右踝关节角度 left_wrist_angles = [] # 左腕关节角度 right_wrist_angles = [] # 右腕关节角度

    存储关键点数据

    keypoints_data = [] # 每一帧的关键点数据

    def calculate_angle(a, b, c): """ 计算三个点之间的夹角(以度为单位)。 a, b, c 是三个点的坐标,b 是顶点。 """ a = np.array(a) b = np.array(b) c = np.array(c)

    # 计算向量
    ba = a - b
    bc = c - b
    
    # 计算夹角
    cosine_angle = np.dot(ba, bc) / (np.linalg.norm(ba) * np.linalg.norm(bc))
    angle = np.arccos(cosine_angle)
    return np.degrees(angle)
    

    使用本地视频推理

    cap = cv2.VideoCapture(r"G:\运动方案\动作评价\卧推.mp4") with mp_pose.Pose( min_detection_confidence=0.5, min_tracking_confidence=0.5, model_complexity=1) as pose:

    frame_count = 0
    while cap.isOpened():
        success, image = cap.read()
        if not success:
            print("Video ended or corrupted.")
            break  # 视频结束或损坏时退出
    
        # 镜像翻转视频帧
        image = cv2.flip(image, 1)  # 水平翻转图像
    
        start = time.time()
    
        # 提高性能
        image.flags.writeable = False
        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
        results = pose.process(image)
    
        # 绘制姿态注释
        image.flags.writeable = True
        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    
        mp_drawing.draw_landmarks(
            image,
            results.pose_landmarks,
            mp_pose.POSE_CONNECTIONS,
            landmark_drawing_spec=mp_drawing_styles.get_default_pose_landmarks_style())
    
        end = time.time()
        fps = 1 / (end - start)
        fps_text = f"FPS: {fps:.2f}"
        # 实时显示帧数
        cv2.putText(image, fps_text, (100, 50),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.75, (255, 255, 255), 3)
    
        # 如果检测到世界坐标中的关键点,计算关节角度
        if results.pose_world_landmarks:
            # 获取关键点数据
            frame_keypoints = []
            for landmark in results.pose_world_landmarks.landmark:
                frame_keypoints.append([landmark.x, landmark.y, landmark.z])
            keypoints_data.append(frame_keypoints)
    
            # 计算左肩关节角度
            left_shoulder = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].x,
                             results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].y,
                             results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].z]
            left_elbow = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].x,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].y,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].z]
            left_hip = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].x,
                        results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].y,
                        results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].z]
    
            left_shoulder_angle = calculate_angle(left_elbow, left_shoulder, left_hip)
            left_shoulder_angles.append(left_shoulder_angle)
    
            # 计算右肩关节角度
            right_shoulder = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].x,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].y,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].z]
            right_elbow = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].x,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].y,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].z]
            right_hip = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].x,
                         results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].y,
                         results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].z]
    
            right_shoulder_angle = calculate_angle(right_elbow, right_shoulder, right_hip)
            right_shoulder_angles.append(right_shoulder_angle)
    
            # 计算左肘关节角度
            left_wrist = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].x,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].y,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].z]
    
            left_elbow_angle = calculate_angle(left_shoulder, left_elbow, left_wrist)
            left_elbow_angles.append(left_elbow_angle)
    
            # 计算右肘关节角度
            right_wrist = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].x,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].y,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].z]
    
            right_elbow_angle = calculate_angle(right_shoulder, right_elbow, right_wrist)
            right_elbow_angles.append(right_elbow_angle)
    
            # 计算左髋关节角度
            left_knee = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_KNEE].x,
                         results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_KNEE].y,
                         results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_KNEE].z]
    
            left_hip_angle = calculate_angle(left_shoulder, left_hip, left_knee)
            left_hip_angles.append(left_hip_angle)
    
            # 计算右髋关节角度
            right_knee = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_KNEE].x,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_KNEE].y,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_KNEE].z]
    
            right_hip_angle = calculate_angle(right_shoulder, right_hip, right_knee)
            right_hip_angles.append(right_hip_angle)
    
            # 计算左膝关节角度
            left_ankle = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ANKLE].x,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ANKLE].y,
                          results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ANKLE].z]
    
            left_knee_angle = calculate_angle(left_hip, left_knee, left_ankle)
            left_knee_angles.append(left_knee_angle)
    
            # 计算右膝关节角度
            right_ankle = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ANKLE].x,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ANKLE].y,
                           results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ANKLE].z]
    
            right_knee_angle = calculate_angle(right_hip, right_knee, right_ankle)
            right_knee_angles.append(right_knee_angle)
    
            # 计算左踝关节角度
            left_foot_index = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].x,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].y,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_FOOT_INDEX].z]
    
            left_ankle_angle = calculate_angle(left_knee, left_ankle, left_foot_index)
            left_ankle_angles.append(left_ankle_angle)
    
            # 计算右踝关节角度
            right_foot_index = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].x,
                                results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].y,
                                results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_FOOT_INDEX].z]
    
            right_ankle_angle = calculate_angle(right_knee, right_ankle, right_foot_index)
            right_ankle_angles.append(right_ankle_angle)
    
            # 计算左腕关节角度
            left_wrist_angle = calculate_angle(left_elbow, left_wrist, left_shoulder)
            left_wrist_angles.append(left_wrist_angle)
    
            # 计算右腕关节角度
            right_wrist_angle = calculate_angle(right_elbow, right_wrist, right_shoulder)
            right_wrist_angles.append(right_wrist_angle)
    
            # 输出角度数据
            print(f"Frame {frame_count}:")
            print(f"Left Shoulder Angle: {left_shoulder_angle:.2f} degrees")
            print(f"Right Shoulder Angle: {right_shoulder_angle:.2f} degrees")
            print(f"Left Elbow Angle: {left_elbow_angle:.2f} degrees")
            print(f"Right Elbow Angle: {right_elbow_angle:.2f} degrees")
            print(f"Left Hip Angle: {left_hip_angle:.2f} degrees")
            print(f"Right Hip Angle: {right_hip_angle:.2f} degrees")
            print(f"Left Knee Angle: {left_knee_angle:.2f} degrees")
            print(f"Right Knee Angle: {right_knee_angle:.2f} degrees")
            print(f"Left Ankle Angle: {left_ankle_angle:.2f} degrees")
            print(f"Right Ankle Angle: {right_ankle_angle:.2f} degrees")
            print(f"Left Wrist Angle: {left_wrist_angle:.2f} degrees")
            print(f"Right Wrist Angle: {right_wrist_angle:.2f} degrees")
            print("-" * 40)
    
        # 显示视频帧
        cv2.imshow('MediaPipe Pose', image)
        if cv2.waitKey(5) & 0xFF == 27:
            break
    
        frame_count += 1
    

    cap.release() cv2.destroyAllWindows()

    保存数据为 .npz 文件

    output_npz_path = r"C:\Users\messi10\Desktop\运动方案\动作评价\八段锦\wotui.npz"

    np.savez(

    output_npz_path,

    left_shoulder_angles=left_shoulder_angles,

    right_shoulder_angles=right_shoulder_angles,

    left_elbow_angles=left_elbow_angles,

    right_elbow_angles=right_elbow_angles,

    left_hip_angles=left_hip_angles,

    right_hip_angles=right_hip_angles,

    left_knee_angles=left_knee_angles,

    right_knee_angles=right_knee_angles,

    left_ankle_angles=left_ankle_angles,

    right_ankle_angles=right_ankle_angles,

    left_wrist_angles=left_wrist_angles,

    right_wrist_angles=right_wrist_angles,

    keypoints_data=keypoints_data

    )

    print(f"Joint data saved to: {output_npz_path}")

    保存数据为 .csv 文件

    output_csv_path = r"F:\pythonzfy\动作评价\velocity\wotui.csv" with open(output_csv_path, mode='w', newline='') as file: writer = csv.writer(file) # 写入表头 writer.writerow([ "Frame", "Left Shoulder Angle", "Right Shoulder Angle", "Left Elbow Angle", "Right Elbow Angle", "Left Hip Angle", "Right Hip Angle", "Left Knee Angle", "Right Knee Angle", "Left Ankle Angle", "Right Ankle Angle", "Left Wrist Angle", "Right Wrist Angle", "Keypoints" ]) # 写入每一帧的数据 for i in range(len(left_shoulder_angles)): writer.writerow([ i, left_shoulder_angles[i], right_shoulder_angles[i], left_elbow_angles[i], right_elbow_angles[i], left_hip_angles[i], right_hip_angles[i], left_knee_angles[i], right_knee_angles[i], left_ankle_angles[i], right_ankle_angles[i], left_wrist_angles[i], right_wrist_angles[i], keypoints_data[i] ]) print(f"Joint data saved to: {output_csv_path}")改成只要肩肘腕关节的数据

  2. Qwen Assistant

    你希望将代码修改为 只保留肩、肘、腕三个关节的角度数据,并相应地更新输出文件(.csv)中的内容。下面是针对你的需求所做的修改说明和完整代码。


    ✅ 修改目标:

    1. 删除所有髋、膝、踝关节角度的计算与存储
    2. 仅保留左/右肩、肘、腕三个关节角度
    3. 更新 .csv 文件格式以匹配新数据结构

    ✅ 修改后的代码如下:

    import cv2
    import mediapipe as mp
    import numpy as np
    import time
    import csv
    
    # 初始化MediaPipe Pose模块
    mp_pose = mp.solutions.pose
    mp_drawing = mp.solutions.drawing_utils
    mp_drawing_styles = mp.solutions.drawing_styles
    
    # 存储角度数据(只保留肩、肘、腕)
    left_shoulder_angles = []
    right_shoulder_angles = []
    left_elbow_angles = []
    right_elbow_angles = []
    left_wrist_angles = []
    right_wrist_angles = []
    
    # 存储关键点数据
    keypoints_data = []  # 每一帧的关键点数据
    
    def calculate_angle(a, b, c):
        """
        计算三个点之间的夹角(以度为单位)。
        a, b, c 是三个点的坐标,b 是顶点。
        """
        a = np.array(a)
        b = np.array(b)
        c = np.array(c)
    
        ba = a - b
        bc = c - b
    
        cosine_angle = np.dot(ba, bc) / (np.linalg.norm(ba) * np.linalg.norm(bc))
        angle = np.arccos(cosine_angle)
        return np.degrees(angle)
    
    # 使用本地视频推理
    cap = cv2.VideoCapture(r"G:\运动方案\动作评价\卧推.mp4")
    with mp_pose.Pose(
            min_detection_confidence=0.5,
            min_tracking_confidence=0.5,
            model_complexity=1) as pose:
    
        frame_count = 0
        while cap.isOpened():
            success, image = cap.read()
            if not success:
                print("Video ended or corrupted.")
                break  # 视频结束或损坏时退出
    
            # 镜像翻转视频帧
            image = cv2.flip(image, 1)  # 水平翻转图像
    
            start = time.time()
    
            # 提高性能
            image.flags.writeable = False
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            results = pose.process(image)
    
            # 绘制姿态注释
            image.flags.writeable = True
            image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    
            mp_drawing.draw_landmarks(
                image,
                results.pose_landmarks,
                mp_pose.POSE_CONNECTIONS,
                landmark_drawing_spec=mp_drawing_styles.get_default_pose_landmarks_style())
    
            end = time.time()
            fps = 1 / (end - start)
            fps_text = f"FPS: {fps:.2f}"
            # 实时显示帧数
            cv2.putText(image, fps_text, (100, 50),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.75, (255, 255, 255), 3)
    
            # 如果检测到世界坐标中的关键点,计算关节角度
            if results.pose_world_landmarks:
                # 获取关键点数据
                frame_keypoints = []
                for landmark in results.pose_world_landmarks.landmark:
                    frame_keypoints.append([landmark.x, landmark.y, landmark.z])
                keypoints_data.append(frame_keypoints)
    
                # 左肩角度
                left_shoulder = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].x,
                                 results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].y,
                                 results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].z]
                left_elbow = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].x,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].y,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].z]
                left_hip = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].x,
                            results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].y,
                            results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].z]
                left_shoulder_angle = calculate_angle(left_elbow, left_shoulder, left_hip)
                left_shoulder_angles.append(left_shoulder_angle)
    
                # 右肩角度
                right_shoulder = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].x,
                                  results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].y,
                                  results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].z]
                right_elbow = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].x,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].y,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].z]
                right_hip = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].x,
                             results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].y,
                             results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].z]
                right_shoulder_angle = calculate_angle(right_elbow, right_shoulder, right_hip)
                right_shoulder_angles.append(right_shoulder_angle)
    
                # 左肘角度
                left_wrist = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].x,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].y,
                              results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].z]
                left_elbow_angle = calculate_angle(left_shoulder, left_elbow, left_wrist)
                left_elbow_angles.append(left_elbow_angle)
    
                # 右肘角度
                right_wrist = [results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].x,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].y,
                               results.pose_world_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].z]
                right_elbow_angle = calculate_angle(right_shoulder, right_elbow, right_wrist)
                right_elbow_angles.append(right_elbow_angle)
    
                # 左腕角度
                left_wrist_angle = calculate_angle(left_elbow, left_wrist, left_shoulder)
                left_wrist_angles.append(left_wrist_angle)
    
                # 右腕角度
                right_wrist_angle = calculate_angle(right_elbow, right_wrist, right_shoulder)
                right_wrist_angles.append(right_wrist_angle)
    
                # 输出角度数据
                print(f"Frame {frame_count}:")
                print(f"Left Shoulder Angle: {left_shoulder_angle:.2f} degrees")
                print(f"Right Shoulder Angle: {right_shoulder_angle:.2f} degrees")
                print(f"Left Elbow Angle: {left_elbow_angle:.2f} degrees")
                print(f"Right Elbow Angle: {right_elbow_angle:.2f} degrees")
                print(f"Left Wrist Angle: {left_wrist_angle:.2f} degrees")
                print(f"Right Wrist Angle: {right_wrist_angle:.2f} degrees")
                print("-" * 40)
    
            # 显示视频帧
            cv2.imshow('MediaPipe Pose', image)
            if cv2.waitKey(5) & 0xFF == 27:
                break
    
            frame_count += 1
    
    cap.release()
    cv2.destroyAllWindows()
    
    # 保存数据为 .csv 文件
    output_csv_path = r"F:\pythonzfy\动作评价\velocity\wotui.csv"
    with open(output_csv_path, mode='w', newline='') as file:
        writer = csv.writer(file)
        # 写入表头
        writer.writerow([
            "Frame",
            "Left Shoulder Angle", "Right Shoulder Angle",
            "Left Elbow Angle", "Right Elbow Angle",
            "Left Wrist Angle", "Right Wrist Angle",
            "Keypoints"
        ])
        # 写入每一帧的数据
        for i in range(len(left_shoulder_angles)):
            writer.writerow([
                i,
                left_shoulder_angles[i], right_shoulder_angles[i],
                left_elbow_angles[i], right_elbow_angles[i],
                left_wrist_angles[i], right_wrist_angles[i],
                keypoints_data[i]  # 关键点数据仍保留(可选)
            ])
    
    print(f"Joint data saved to: {output_csv_path}")
    

    ✅ 输出示例(CSV 表头):

    FrameLeft Shoulder AngleRight Shoulder AngleLeft Elbow AngleRight Elbow AngleLeft Wrist AngleRight Wrist AngleKeypoints
    098.6597.32165.43164.1222.123.5[[x,y,z],...]

    如需进一步简化(例如去掉 keypoints_data),也可以移除相关变量与写入逻辑。欢迎继续提问!

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.