Human body key point labeling is a computer vision task, which refers to manually marking key points at specified positions, such as face feature points, human bone connection points, etc., which are often used to train facial recognition models and statistical models. These keypoints can represent various aspects of the image, such as corners, edges, or specific features. In facial recognition, they can label eyes, noses, and mouths, and in human pose recognition, keypoints can represent body joints.
Human body keypoint annotation can also be used to analyze the spatial relationship between multiple objects, such as football players on the pitch. By marking the key points of football players, information such as the position, posture, and trajectory of each player can be obtained, and the overall tactics and coordination of the team can be further analyzed. This method can be applied to team sports such as football, basketball, and ice hockey to provide data support for team tactical analysis and decision-making.
Specific application scenarios for human body key point labeling include but are not limited to:
- Facial recognition: mark eyes, nose, mouth and other facial feature points for applications such as face recognition and expression recognition.
- Human body posture recognition: Marking the joints of the human body can be used to recognize the posture of the human body, including the position and direction of the head, arms, legs, etc., and can be applied to athlete training, medical rehabilitation, game interaction and other fields.
- Behavior recognition: mark human body actions and behaviors, such as applications in security monitoring, intelligent transportation and other fields.
- Medical image analysis: mark key points such as human organs and diseased parts in medical images for disease diagnosis and treatment.
When labeling key points of the human body, it usually requires the help of professional tools or machine learning models. The specific process includes steps such as labeling images, selecting labeling tools, labeling key points, and adjusting labeling results. At the same time, in order to improve the accuracy and consistency of labeling, manual review and correction may be required.
The key point labeling process requires professional knowledge and skills, and it also takes a lot of time and energy. In contrast, bounding boxes and polygon annotations are usually easier to annotate and are suitable for some basic computer vision tasks such as object detection.
However, for some specific tasks and scenarios, bounding box and polygon annotations may not provide enough detailed information. For example, in human posture recognition and behavior analysis, key point annotation can provide more accurate information, including the position of each joint of the human body, posture angle, etc., which is very important for understanding the movement and behavior of the human body. Therefore, in some cases, it is necessary to use keypoint annotations.
The main difficulties in human body key point labeling include:
- Occlusion problem: Some parts of the human body may be occluded by other objects, such as arms, legs, etc., which will bring certain difficulties to the labeling of key points.
- Pose diversity: Human body poses are ever-changing, and the positions of key points under different poses are also different, which brings great challenges to labeling.
- Algorithm robustness: human body key point labeling requires high algorithm robustness, because human body posture and environmental conditions will change with time, space and other factors, and the algorithm needs to be able to adapt to different scenarios.
- Insufficient labeled data: In some fields, such as medical care and smart clothing, there are relatively few labeled datasets, which will bring certain difficulties to model training and evaluation.
In order to solve the problem of insufficient labeling data, JLW Technology has developed its own data labeling platform. The labeling workbench is equipped with intelligent auxiliary labeling functions to improve labeling efficiency, and supports automatic labeling of image object content, with an accuracy rate of up to 97%. Manual intervention into modification can reduce the burden of manual labeling to a certain extent and improve labeling efficiency and accuracy.
JLW Technology attaches great importance to the daily skill training of labeling and quality inspection personnel. The quality inspection personnel are trained by the project supervisor to ensure that the training content is closely related to the actual work. Quality inspection is carried out at the same time as labeling, and the quality inspection is carried out in real time. Find and solve problems in a timely manner, continuously improve the training effect, and ensure the effectiveness of the training. All data are subject to sampling inspection by the team leader, inspection by quality inspectors, and inspection by the project manager for final delivery.
As a leading AI basic data company, Jinglianwen Technology has rich experience in implementing AI data projects and a complete project management integration process. In the entire life cycle of AI data services, it adopts advanced technology and management strategies to meet AI data processing requires real-time, accuracy and security.
After years of accumulation, we have 5 labeling bases across the country and thousands of full-time labelers. The intelligent labeling platform covers most of the mainstream labeling tools, supports the labeling of key information points of faces and human bodies, and can increase the accuracy of models by 30%. % or more, the iteration cycle is greatly shortened, and the cost of a single model training can be saved by 30%. It can also provide customized services for enterprises according to the actual needs of enterprises, and achieve a significant improvement in the effect of large-scale implementation of AI applications.
The detailed semantic definition and description of the 25 key points of the human body are as follows.
Described as follows:
| serial number |
point position definition |
| 0 |
Nose tip position |
| 1 |
Neck point, the center of the line connecting the centers of the left and right shoulder joints |
| 2 |
right shoulder center |
| 3 |
center of right elbow |
| 4 |
center of right wrist |
| 5 |
left shoulder center |
| 6 |
center of left elbow |
| 7 |
center of left wrist |
| 8 |
The center of the line connecting the centers of the hip joints of the left and right legs |
| 9 |
right leg hip center |
| 10 |
Center of right knee joint |
| 11 |
At the center of the ankle joint of the right leg |
| 12 |
left leg hip center |
| 13 |
Center of left knee joint |
| 14 |
At the center of the ankle joint of the left leg |
| 15 |
center of right eye |
| 16 |
center of left eye |
| 17 |
right ear hole |
| 18 |
left ear hole |
| 19 |
center of left thumb |
| 20 |
Left pinky center |
| 21 |
left heel |
| 22 |
center of right thumb |
| 23 |
Right pinky center |
| 24 |
right heel |
5 points on the head, including left ear, left eye, nose, right eye, and right ear; 7 points on the upper body, including left wrist, left elbow, left shoulder, right shoulder, right elbow, right wrist, and neck; 7 points on the lower body , including the left ankle, left knee, 2 points on both sides of the thigh (not too edge), the center of the thigh, the right knee, and the right ankle; 6 points on the foot, including the left heel, left big toe, left little toe, Right heel, right big toe, right little toe.
JLW Technology｜Data Collection｜Data Labeling
Helping artificial intelligence technology, empowering the intelligent transformation and upgrading of traditional industries
The copyright of the text and graphics of the article belongs to Jinglianwen Technology. For commercial reprinting, please contact Jinglianwen Technology for authorization. For non-commercial reprinting, please indicate the source.