single_human_pose
    ├─data
    │  ├─game                           // 体感游戏资源目录
    │      ├─icon                       // 体感游戏所需的图标资源
    │      ├─src // 体感游戏源代码，可独立运行
    │  └─dance_120fps.mp4               // 测试视频
    ├─etc
    │  └─flowunit // 功能单元目录
    │      ├─draw_pose // 关键点绘制功能单元
    │         ├─draw_pose.py            // 关键点绘制功能单元入口文件
    │         ├─draw_pose.toml // 关键点绘制功能单元配置文件
    │         ├─draw_utils.py           // 其他功能函数存放文件
    │         ├─flappy.py               // Flappy Bird核心逻辑
    │         ├─smooth.py               // 平滑算法，使关键点更稳定
    │         └─vector.py               // 平面坐标系中点的运算
    │      ├─... // 其他功能单元
    ├─graph
    │  ├─single_human_pose.toml // 默认的技能流程图
    │  └─single_human_pose_camera.toml // 使用摄像头运行的技能流程图
    ├─...
    └─build_project.sh

def run():
    # 初始化游戏
    snake = Snake()    # 贪吃蛇
    flappy = Flappy()  # Flappy Bird
 pacman = Pacman()  # 吃豆人
    # 初始化摄像头，参数0表示设备的第一个摄像头
    cap = cv2.VideoCapture(0)
    # 判断摄像头初始化是否成功
 if not cap.isOpened():
 print('failed to open camera 0')
 exit()
    # 设置分辨率为 960 x 540
 cap.set(3, 960)
 cap.set(4, 540)
 while True:
        # 读取一帧视频图像，ret表示读取是否成功
        ret, frame = cap.read()
        # 对原始图像做高斯模糊，避免干扰到游戏画面
        frame = cv2.GaussianBlur(frame, (0, 0), 5)
        # 阻塞等待键盘响应1ms，获取按下的按键值
 pressed_key = cv2.waitKey(1) & 0xFF
 if pressed_key == ord('q'): # 如果按下q键则退出游戏
 break
 else: # 根据按键类型更新游戏画面
            # frame = snake.update_snake_keyboard(frame, pressed_key)
            frame = flappy.update_flappy_keyboard(frame, pressed_key)
            # frame = pacman.update_pacman_keyboard(frame, pressed_key)
        # 打开一个名为game的窗口，显示图像
        cv2.imshow('game', frame)
    # 释放摄像头资源
 cap.release()
    # 关闭所有窗口
    cv2.destroyAllWindows()

    def open(self, config):
 ...
        # 使用图标资源初始化Flappy Bird游戏控制示例
 icon_path = config.get_string("icon_path", ".")
 self.flappy = Flappy(icon_path)
 return modelbox.Status.StatusCode.STATUS_SUCCESS
    def process(self, data_context):
 ...
 for image, hand_pose in zip(in_image, in_feat):
 ...
                # 获取上一功能单元输出的人体关键点数据
 pose_data = np.array(hand_pose.as_object(), copy=False)
 pose_data = pose_data.reshape((self.kps, 3))
                # 计算双臂与身体的夹角
 arm_angles = get_arm_angles(bbox, pose_data, self.keypoints_smooth)
 ...
                # 在摄像头画面中画出主要的人体关节，并作高斯模糊
 draw_pose(out_img, bbox, pose_data, self.score_thre)
 out_img = cv2.GaussianBlur(out_img, (0, 0), 5)
                # 根据双臂动作控制游戏画面更新，叠加到摄像头画面中做展示
 out_img, alive = self.flappy.update_flappy_pose(out_img, arm_angles, fps)
 ...
 return modelbox.Status.StatusCode.STATUS_SUCCESS

graphconf = """digraph single_human_pose {
    node [shape=Mrecord]
 queue_size = 1
 batch_size = 1
    input1[type=input,flowunit=input,device=cpu,deviceid=0]
 data_source_parser[type=flowunit, flowunit=data_source_parser, device=cpu, deviceid=0]
 local_camera[type=flowunit, flowunit=local_camera, device=cpu, deviceid=0, pix_fmt=bgr, cam_width=960, cam_height=540]
 det_pre[type=flowunit, flowunit=det_pre, device=cpu, deviceid=0]
 color_transpose[type=flowunit flowunit=packed_planar_transpose device=cpu deviceid=0]
 normalize[type=flowunit flowunit=normalize device=cpu deviceid=0 standard_deviation_inverse="0.003921568627451, 0.003921568627451, 0.003921568627451"]
 det_human[type=flowunit, flowunit=det_human, device=cpu, deviceid=0, batch_size=1]
 det_post[type=flowunit, flowunit=det_post, device=cpu, deviceid=0]
 object_tracker[type=flowunit, flowunit=object_tracker, device=cpu, deviceid=0]
 expand_box[type=flowunit, flowunit=expand_box, device=cpu, deviceid=0]
 image_resize[type=flowunit flowunit=resize device=cpu deviceid="0" image_width=192, image_height=256]
    color_transpose2[type=flowunit flowunit=packed_planar_transpose device=cpu deviceid=0]
 mean[type=flowunit flowunit=mean device=cpu deviceid="0" mean="116.28,103.53,123.68"]
    normalize2[type=flowunit flowunit=normalize device=cpu deviceid="0" standard_deviation_inverse="0.0175070,0.01742919,0.01712475"]
 det_pose[type=flowunit, flowunit=det_pose, device=cpu, deviceid=0, batch_size=1]
 pose_post[type=flowunit, flowunit=pose_post, device=cpu, deviceid=0]
 draw_pose[type=flowunit, flowunit=draw_pose, device=cpu, deviceid=0]
 video_out[type=flowunit, flowunit=video_out, device=cpu, deviceid=0]
    input1 -> data_source_parser:in_data
 data_source_parser:out_video_url -> local_camera:in_camera_packet
 local_camera:out_camera_frame -> det_pre:in_image
 det_pre:resized_image -> color_transpose:in_image
 color_transpose:out_image -> normalize:in_data
 normalize:out_data -> det_human:input
    det_human:output1 -> det_post:in_feat1
    det_human:output2 -> det_post:in_feat2
    det_human:output3 -> det_post:in_feat3
 det_pre:out_image -> det_post:in_image
 det_post:has_human -> object_tracker:in_image
 object_tracker:out_image -> expand_box:in_image
 expand_box:out_image -> image_resize:in_image
 image_resize:out_image -> color_transpose2:in_image
    color_transpose2:out_image -> mean:in_data
 mean:out_data -> normalize2:in_data
    normalize2:out_data -> det_pose:image
 det_pose:heatmap -> pose_post:in_feat
 pose_post:out_data -> draw_pose:in_feat
 object_tracker:out_image -> draw_pose:in_image
 draw_pose:out_image -> video_out:in_video_frame
 det_post:no_human -> video_out:in_video_frame
}"""

[input]
type = "url"
url = "0"  # 表示0号摄像头，即PC自带摄像头，若PC无摄像头需外接USB摄像头
[output]
type = "local"
url = "0:pose_game"  # 表示名为```pose_game```的本地窗口