app.py

import cv2
import mediapipe as mp
import time
import matplotlib.pyplot as plt
import math
import numpy as np
import pandas as pd
from moviepy.video.io.bindings import mplfig_to_npimage
import os

plt.rcParams['font.sans-serif'] = ['SimHei']
# 导入姿态跟踪方法
mpPose = mp.solutions.pose  # 姿态识别方法
pose = mpPose.Pose(static_image_mode=False, # 静态图模式，False代表置信度高时继续跟踪，True代表实时跟踪检测新的结果
                   #upper_body_only=False,  # 是否只检测上半身
                   smooth_landmarks=True,  # 平滑，一般为True
                   min_detection_confidence=0.5, # 检测置信度
                   min_tracking_confidence=0.5)  # 跟踪置信度
# 检测置信度大于0.5代表检测到了，若此时跟踪置信度大于0.5就继续跟踪，小于就沿用上一次，避免一次又一次重复使用模型
 
# 导入绘图方法
mpDraw = mp.solutions.drawing_utils

# 定义一个处理视频的函数
def video_processor(video):
    
    # 创建一个VideoCapture对象
    cap = cv2.VideoCapture(video)

    number_of_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    fps = cap.get(cv2.CAP_PROP_FPS)
    
    
    # 创建一个VideoWriter对象，输出为mp4格式
    fourcc = cv2.VideoWriter_fourcc(*'avc1')
    out = cv2.VideoWriter('output.mp4', fourcc, fps, (width, height))

    # 创建一个图形对象
    fig, ax = plt.subplots(figsize=(3, 3))

    # ax.set_position((0.1, 0.1, 0.8, 0.4))
    ax.set_xlim(0, number_of_frames)
    ax.set_ylim(-10, 200)
    ax.set_title('左腿夹角')
    ax.set_xlabel('Time')
    ax.set_ylabel('Angle')
    line, = ax.plot([], [], 'r-')

    # 定义更新曲线图的函数
    def update_graph(frame, angle):
        # 获取当前时间
        time = frame
        # 添加新的数据点
        line.set_xdata(np.append(line.get_xdata(), time))
        line.set_ydata(np.append(line.get_ydata(), angle))
        # 返回更新后的曲线
        return line

    # 处理每一帧图像
    # lmlist = [] # 存放人体关键点信息
    left_23 = [] # 存 左髋关节的点
    left_25 = [] # 存 左髋关节的点
    a = [] # 存着所有角度

    def angle(p1, p2):
        x1, y1 = p1
        x2, y2 = p2
        dx = x2 - x1
        dy = y2 - y1
        return math.degrees(math.atan2(dy, dx))

    frame_count = 0 # 记录帧数
    while True:
        # 读取一帧图像
        success, img = cap.read()
        if not success:
            print('视频结束')
            break
        frame_count += 1
        # 将图像转为RGB格式
        imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        # 将图像传给姿态识别模型
        results = pose.process(imgRGB)
        # 如果检测到体态就执行下面内容，没检测到就不执行
        if results.pose_landmarks:
            # 绘制姿态坐标点
            mpDraw.draw_landmarks(img, results.pose_landmarks, mpPose.POSE_CONNECTIONS)
            # 获取32个人体关键点坐标, index记录是第几个关键点
            for index, lm in enumerate(results.pose_landmarks.landmark):
                # 保存每帧图像的宽、高、通道数
                h, w, c = img.shape
                # 转换为像素坐标
                cx, cy = int(lm.x * w), int(lm.y * h)
                # 保存坐标信息
                # lmlist.append((cx, cy))
                if index == 23:
                    left_23.append([cx, cy])
                if index == 25:
                    left_25.append([cx, cy])
                # 在关键点上画圆圈
                cv2.circle(img, (cx,cy), 3, (0,255,0), cv2.FILLED)
        # 计算左腿夹角
        if len(left_23) != 0 and len(left_25) != 0:
            a.append(angle(left_23[-1],left_25[-1]))
            # 更新曲线图
            update_graph(frame_count, a[-1])
        # 将图像转换为numpy数组
        graph = mplfig_to_npimage(fig)
        # 将曲线图叠加到视频上
        img[-graph.shape[0]:,-graph.shape[1]:,:] = graph
        # 写入新的视频文件
        out.write(img)
    # 释放资源
    out.release()
    cap.release()
    # 返回输出视频的路径
   
    return 'output.mp4'

import gradio as gr

demo = gr.Interface(fn=video_processor, title='滑板招式姿态识别',
                    inputs=gr.Video(source="upload"), outputs=gr.Video()
                    # examples=["Ollie120.mp4"]
                    )
demo.launch()