tf-pose-estimation 姿态估计实测

墨蓝 2024-04-17 22:36 74阅读 0赞

在研究行为检测时，对tf-pose-estimation进行了测试，它本身比较轻巧，可以利用CPU进行实时的检测，容易跑通。  
我在Ubuntu16.04上进行了配置和测试，记录如下：  
tf-pose-estimation的github地址为：[https://github.com/ildoonet/tf-pose-estimation][https_github.com_ildoonet_tf-pose-estimation]

tf-pose-estimation依赖项：

*  python3
 *  tensorflow 1.4.1+
 *  opencv3, protobuf, python3-tk
 *  slidingwindow

## 安装： ##

1.  下载代码并安装第三方依赖库：

git clone https://www.github.com/ildoonet/tf-pose-estimation
    cd tf-pose-estimation
    pip3 install -r requirements.txt

1.  编译C++库：

cd tf_pose/pafprocess
    swig -python -c++ pafprocess.i && python3 setup.py build_ext --inplace

## 模型下载： ##

在运行前，需要下载模型文件。 您可以在移动设备或其他平台上部署此模型文件。

*  cmu (trained in 656x368)
 *  mobilenet\_thin (trained in 432x368)
 *  mobilenet\_v2\_large (trained in 432x368)
 *  mobilenet\_v2\_small (trained in 432x368)

cd models/graph/cmu
    bash download.sh

## 测试： ##

1.  **单个图片测试**

python run.py --model=mobilenet_thin --resize=432x368 --image=./images/apink3.jpg

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3U0NDExNzgx_size_16_color_FFFFFF_t_70]

1.  **电脑摄像头测试**

python run_webcam.py --model=mobilenet_thin --resize=432x368 --camera=0

1.  **视频测试**
    
    作者没有给出命令，不过自己对run\_video.py稍加修改代码即可，我在抖音上下载了一段视频，进行了测试：  
                                          ![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3U0NDExNzgx_size_16_color_FFFFFF_t_70 1]  
    鉴于有同学不知道如何更改读取视频和保存，我把改好以后的run\_video.py代码贴出来，如下：

import argparse
    import logging
    import time
    
    import cv2
    import numpy as np
    
    from tf_pose import common
    from tf_pose.estimator import TfPoseEstimator
    from tf_pose.networks import get_graph_path, model_wh
    
    logger = logging.getLogger('TfPoseEstimator-Video')
    logger.setLevel(logging.DEBUG)
    ch = logging.StreamHandler()
    ch.setLevel(logging.DEBUG)
    formatter = logging.Formatter('[%(asctime)s] [%(name)s] [%(levelname)s] %(message)s')
    ch.setFormatter(formatter)
    logger.addHandler(ch)
    
    fps_time = 0
    
    
    if __name__ == '__main__':
        parser = argparse.ArgumentParser(description='tf-pose-estimation Video')
        parser.add_argument('--video', type=str, default='./images/crowd.mp4')  #改为自己的输入视频
        parser.add_argument('--resolution', type=str, default='432x368', help='network input resolution. default=432x368')
        parser.add_argument('--model', type=str, default='mobilenet_thin', help='cmu / mobilenet_thin / mobilenet_v2_large / mobilenet_v2_small')
        parser.add_argument('--resize-out-ratio', type=float, default=4.0,help='if provided, resize heatmaps before they are post-processed. default=1.0')
        parser.add_argument('--show-process', type=bool, default=False,help='for debug purpose, if enabled, speed for inference is dropped.')
        parser.add_argument('--showBG', type=bool, default=True, help='False to show skeleton only.')
        args = parser.parse_args()
    
        logger.debug('initialization %s : %s' % (args.model, get_graph_path(args.model)))
        w, h = model_wh(args.resolution)
        e = TfPoseEstimator(get_graph_path(args.model), target_size=(w, h))
        cap = cv2.VideoCapture(args.video)
    
        ret_val, image = cap.read()
        video_size = (image.shape[1],image.shape[0])
        fourcc = cv2.VideoWriter_fourcc('M', 'P', '4', '2')
        #video_size = (2560,1920)
        outVideo = cv2.VideoWriter('save.avi', fourcc, 10, video_size)
        filename = '/home/yasin/save.avi'    #更改为自己的保存路径
        if cap.isOpened() is False:
            print("Error opening video stream or file")
        while cap.isOpened():
            ret_val, image = cap.read()
    
            if not ret_val:
                break
            logger.debug('image process+')
            #image = common.read_imgfile(args.image, None, None)
            humans = e.inference(image, resize_to_default=(w > 0 and h > 0), upsample_size=args.resize_out_ratio)
    
            logger.debug('postprocess+')
            image = TfPoseEstimator.draw_humans(image, humans, imgcopy=False)
            logger.debug('show+')
            # image_res = cv2.resize(image,(640,480))
            cv2.putText(image, "FPS: %f" % (1.0 / (time.time() - fps_time)), (10, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5,(0, 255, 0), 2)
            cv2.imshow('tf-pose-estimation result', image)
            fps_time = time.time()
            outVideo.write(image)
            # video_write = cv2.VideoWriter(filename, fourcc, 10, video_size)
            # video_write.write(image)
    
            if cv2.waitKey(1) == 27:
                break
            logger.debug('finished+')
        cap.release()
        outVideo.release()
        cv2.destroyAllWindows()
        ''' while cap.isOpened(): ret_val, image = cap.read() humans = e.inference(image,resize_to_default=(w > 0 and h > 0)) if not args.showBG: image = np.zeros(image.shape) image = TfPoseEstimator.draw_humans(image, humans, imgcopy=False) cv2.putText(image, "FPS: %f" % (1.0 / (time.time() - fps_time)), (10, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2) cv2.imshow('tf-pose-estimation result', image) fps_time = time.time() if cv2.waitKey(1) == 27: break cv2.destroyAllWindows() logger.debug('finished+') '''

[https_github.com_ildoonet_tf-pose-estimation]: https://github.com/ildoonet/tf-pose-estimation
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3U0NDExNzgx_size_16_color_FFFFFF_t_70]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/73bb901f2f5f4bc49d5a7a7069fbe7f8.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3U0NDExNzgx_size_16_color_FFFFFF_t_70 1]: https://image.dandelioncloud.cn/pgy_files/images/2024/04/17/6b06762a281f4990a629e2299c98c4c9.png