A Greedy Part Assignment Algorithm for Real-Time Multi-person 2D Pose Estimation

Abstract

Human pose-estimation in a multi-person image involves detection of various body parts and grouping them into individual person clusters. While the former task is challenging due to mutual occlusions, the combinatorial complexity of the latter task is very high. We propose a greedy part assignment algorithm that exploits the inherent structure of the human body to achieve a lower complexity, compared to any of the prior published works. This is accomplished by (i) reducing the number of partcandidates using the estimated number of people in the image, (ii) doing a greedy sequential assignment of partclasses, following the kinematic chain from head to ankle (iii) doing a greedy assignment of parts in each part-class set, to person-clusters (iv) limiting the candidate person clusters to the most proximal clusters using human anthropometric data and (v) using only a specific subset of pre-assigned parts for establishing pairwise structural constraints. We show that, these steps sparsify the bodyparts relationship graph and reduces the algorithm's complexity to be linear in the number of candidates of any single part-class. We also propose methods for improving the accuracy of pose-estimation by (i) spawning personclusters from any unassigned significant body part and (ii) suppressing hallucinated parts. On the MPII multi-person pose database, pose-estimation using the proposed method takes only 0.14 seconds per image. We show that, our proposed algorithm, by using a large spatial and structural context, achieves the state-of-the-art accuracy on both MPII and WAF multi-person pose datasets, demonstrating the robustness of our approach.

Learning to Refine Human Pose Estimation

“…We evaluate our approach on four human pose estimation benchmarks, namely MPII Single Person [3], MPII Multi-Person [3], PoseTrack Multi-Person Pose Estimation [2], and PoseTrack Multi-Person Pose Tracking [2]. We report consistent improvement after applying the proposed refinement network to pose predictions given by various stateof-the-art approaches [48,14,23,26,17,46,33,44,6,7] across different datasets and tasks, showing the effectiveness and generality of the proposed framework. With our refinement network, we improve the best reported results for multi-person pose estimation and pose tracking on MPII Human Pose and PoseTrack datasets.…”

“…Initial Pose [46] Refined Pose GT Pose Initial Pose [17] Refined Pose GT Pose Initial Pose [44] Refined Pose GT Pose Initial Pose [6] Refined Pose image. If such a neighbor joint exists in a 75px vicinity, replacement is done with 30% probability.…”

“…Since the output of a multi-person pose estimator is a set of body poses in an image, we can use the PoseRefiner to perform error correction on each estimated pose, independently of the others. Table 2 shows the quantitative effect that the refinement post processing step has on several methods [33,6,23,44] applied on the MPII Multi-Person test set. It proves to help the overall performance of each system, including the best performing method [33] 2.9 mAP) can be considered significant for the localization of joints.…”

High-speed multi-person pose estimation with deep feature transfer

“…Levinkov et al (2017) consider the articulated human body pose estimation as a combinatorial optimisation problem and propose two local search algorithms that offer a feasible solution at given time constraint. Varadarajan et al (2018) exploit the inherent structure of the human body to decrease the complexity of the body part grouping model. Newell et al (2017) propose learning both body part detection and grouping in the CNNs simultaneously.…”

MultiPoseNet: Fast Multi-Person Pose Estimation Using Pose Residual Network

“…Most of the recent approaches use Convolutional Neural Networks (CNNs) to detect body parts and relationships between them in an end-to-end manner [2][3][4]8,18,34], then use assignment algorithms [2][3][4]34] to form individual skeletons.…”