Forest Walk Methods for Localizing Body Joints from Single Depth Image.

Jung HY, Lee S, Heo YS, Yun ID - PLoS ONE (2015)

Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4581738&req=5

pone.0138328.g001: The red lines represent the random forest walks trained to find the position of the head.The random walk starts from the body center in (a). In (b), the head position is found with fewer steps by starting from the chest, which is much closer than the body center.
Mentions: In this paper, we present a simple, yet powerful discriminative method for body part localization from a single depth image, called the Random Tree Walks (RTW) method [13]. The proposed RTW combines random trees and random walk to train the relative directional path to a specific joint from any point. The RTW comprises two stages. The first is to train a random tree (RT) to estimate the relative direction to the joint from a given point. The second is the joint localization stage, where an initial starting point is moved towards the joint position by random walk in the direction estimated from the trained random regression tree. In each step of random tree walk, we are sampling closer to the targeted body joint. Regardless of the size of body part, the number of random walk can be kept minimum and consistent through all body joints. Fig 1 shows an example of the proposed random tree walk process to estimate position of the head. We can see the path of the walk as the regression tree guides the direction of each step at each point. Furthermore, the kinematic tree of joints can be leveraged by performing RTW sequentially, and by initializing the subsequent RTW’s starting point as the estimated position of a preceding joint. We construct an optimal sequence of joints based on the kinematic tree. The comparison between Fig 1a and 1b demonstrates the efficiency of our method in finding the head position.

Bottom Line: A regression forest is trained to estimate the probability distribution to the direction or offset toward the particular joint, relative to the adjacent position.The distribution for next position is found from traversing the regression tree from new position.The continual position sampling through 3D space will eventually produce an expectation of sample positions, which we estimate as the joint position.

View Article: PubMed Central - PubMed

Affiliation: Division of Computer and Electronic Systems Engineering, Hankuk University of Foreign Studies, Yongin, Republic of Korea.

ABSTRACT
We present multiple random forest methods for human pose estimation from single depth images that can operate in very high frame rate. We introduce four algorithms: random forest walk, greedy forest walk, random forest jumps, and greedy forest jumps. The proposed approaches can accurately infer the 3D positions of body joints without additional information such as temporal prior. A regression forest is trained to estimate the probability distribution to the direction or offset toward the particular joint, relative to the adjacent position. During pose estimation, the new position is chosen from a set of representative directions or offsets. The distribution for next position is found from traversing the regression tree from new position. The continual position sampling through 3D space will eventually produce an expectation of sample positions, which we estimate as the joint position. The experiments show that the accuracy is higher than current state-of-the-art pose estimation methods with additional advantage in computation time.

No MeSH data available.

Related in: MedlinePlus