The paper ：DeepCut: Joint Subset Partition and Labeling for Multi Person Pose EstimationĭeepCut It is a bottom-up multi person human posture estimation method. Go through the above steps, The human pose skeleton can be estimated and assigned to each person in the image. Use PAF value, You can trim weak links in bipartite graphs. Use the component confidence graph, A bipartite diagram is formed between component pairs （ As shown in the figure above ）. The subsequent stage is used to refine the prediction made by each branch. Use OpenPose Steps of human posture estimation. This is shown below OpenPose The architecture of the model. Like many bottom-up approaches ,OpenPose First, detect the part belonging to everyone in the image （ Key points ）, Then assign the part to different individuals. OpenPose It is one of the most popular bottom-up human posture estimation methods, Part of the reason is that they are well documented GitHub Implementation code. The paper ：OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields(2019) In the next section, We will review some popular top-down and bottom-up approaches. In this paper, We will focus on multi person pose estimation using deep learning technology. It is difficult to determine which method has better overall performance, Because it actually boils down to people detectors and associations / Which of the grouping algorithms is better. Usually, The top-down approach is easier to implement than the bottom-up approach, Because adding a person detector is better than adding an association / The grouping algorithm is much simpler. This approach is called the bottom-up approach. Īnother method is to detect all parts of the image （ That is, everyone's part ）, And then connect / Groups belong to different people. This approach is called the top-down approach. The simple way is to first combine a human detector, Then estimate the parts, Then calculate everyone's posture. Usually, We can use one of the following two methods to solve the above problem ： Multi person posture estimation is more difficult than single person situation, Because the position and number of people in the image are unknown. Natural, These methods are not particularly useful in many real-life scenes where images contain many people.
These methods usually identify each part first, Then form a connection between them to create a pose. Earliest （ And the slowest ） The method is usually to estimate the pose of a single person in an image with only one person. These years, Several human pose estimation methods have been introduced. Right picture ： Rendered human pose skeleton. Left ： Human posture skeleton COCO Key format.
0 kommentar(er)
