姿态分析开源工具箱MMPose使用示例:人体姿势估计

Posted fengbingchun

tags:

篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了姿态分析开源工具箱MMPose使用示例:人体姿势估计相关的知识,希望对你有一定的参考价值。

      MMPose的介绍及安装参考:https://blog.csdn.net/fengbingchun/article/details/126676309,这里给出人体姿势估计的测试代码,论文:《Deep high-resolution representation learning for human pose estimation》:

      (1).准备测试图像:原始图像来自网络

image_path = "../../data/image/"
image_name = "human.png"

      (2).通过MMDetection模块检测人体框:注,为了去除伪人体框,这里设置了一个阈值,只有大于此阈值的框才作为后面的人体姿势估计

def mmdet_human_detection(device, image, threshold=0.9):
	path = "../../data/model/"
	checkpoint = "faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth"
	url = "https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth"
	download_checkpoint(path, checkpoint, url)

	config = "../../src/mmpose/demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py"
	model = init_detector(config, path+checkpoint, device)

	mmdet_results = inference_detector(model, image)
	# print(mmdet_results)

	human_results = process_mmdet_results(mmdet_results)
	# print(human_results)

	filter_results = []
	mat = cv2.imread(image)
	for result in human_results:
		print("result:", result)
		if result['bbox'][4] > threshold:
			filter_results.append(result)
			cv2.rectangle(mat, (int(result['bbox'][0]), int(result['bbox'][1])), (int(result['bbox'][2]), int(result['bbox'][3])), (255, 0, 0), 1)

	cv2.imwrite("../../data/result_mmpose_2d_human_detection.png", mat)
	cv2.imshow("show", mat)
	cv2.waitKey(0)

	return filter_results

 

      (3).下载人体姿势估计模型: 

def download_checkpoint(path, name, url):
	if os.path.isfile(path+name) == False:
		print("checkpoint(model) file does not exist, now download ...")
		subprocess.run(["wget", "-P", path, url])

path = "../../data/model/"
checkpoint = "hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth"
url = "https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth"
download_checkpoint(path, checkpoint, url)

      (4).根据配置文件和checkpoint文件构建人体估计模型:

config = "../../src/mmpose/configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_256x192.py"
model = init_pose_model(config, path+checkpoint, device)

      (5).进行人体姿势估计推理,输入包括检测到的人体框:

pose_results, returned_outputs = inference_top_down_pose_model(model, image, human_bbox_results, bbox_thr=None, format='xyxy')
print(pose_results)

      (6).显示及保存结果:

vis_pose_result(model, image, pose_results, radius=1, thickness=1, show=True, out_file="../../data/result_mmpose_2d_human_pose_estimation.png")

      执行结果如下图所示: 

      GitHub: https://github.com/fengbingchun/PyTorch_Test

以上是关于姿态分析开源工具箱MMPose使用示例:人体姿势估计的主要内容,如果未能解决你的问题,请参考以下文章

姿态分析开源工具箱MMPose使用示例:2d手势估计

姿态分析开源工具箱MMPose安装及使用示例(2d face landmark detection)

MMPose

MMPose理解

运动人体姿态提取基于GA优化的从二维关节点坐标中估计三维人体姿势的MATLAB仿真

OpenSitUp开源项目:零基础开发基于姿态估计的运动健身APP