Lower Body Pose Estimation in Team Sports Videos Using Label-Grid Classifier Integrated with Tracking-by-Detection

Abstract

Abstract:We propose a human lower body pose estimation method for team sport videos, which is integrated with tracking-by-detection technique. The proposed Label-Grid classifier uses the grid histogram feature of the tracked window from the tracker and estimates the lower body joint position of a specific joint as the class label of the multiclass classifiers, whose classes correspond to the candidate joint positions on the grid. By learning various types of player poses and scales of Histogram-of-Oriented Gradients features within one team sport, our method can estimate poses even if the players are motion-blurred and low-resolution images without requiring a motion-model regression or part-based model, which are popular vision-based human pose estimation techniques. Moreover, our method can estimate poses with part-occlusions and non-upright side poses, which part-detector-based methods find it difficult to estimate with only one model. Experimental results show the advantage of our method for side running poses and non-walking poses. The results also show the robustness of our method for a large variety of poses and scales in team sports videos.

Upper Body Pose Estimation for Team Sports Videos Using a Poselet-Regressor of Spine Pose and Body Orientation Classifiers Conditioned by the Spine Angle Prior

“…Our alignment-based body orientation classification, guided by the 2D spine pose, can predict not only the body orientation but also the 2D spine pose even when hard-occlusions or part disappearance occurs, because it uses a few selected features within the aligned upper body window. This alignment-based pose estimation framework, is suitable for side view running poses as [25] and suitable for upper body bending poses which both frequently appear in team sports videos.…”

“…Since our previous work [16] depends on tracking the whole body, it can only track and estimate the pose of isolated players. Our new design, which tracks the head and uses the head-center-aligned upper body appearance, which is inspired by our another lower body pose estimation framework [25], opens up more chances to track and estimate the pose of players even in congested situations in team sports by only tracking the head region and using only the tracked upper body region for spine pose and body orientation estimation.…”

“…Our tracker-based joint location estimation of the poselet-regressor is inspired by our previous work [25], which also uses the body tracker for the window alignment and used random classification forests to estimate the lower body joint locations.…”

“…Adopting this design of relative joint location estimation by discarding the detection and segmentation ability of the original poselets [15], [22], our poselet-regressor can obtain the relative movement of one joint location from the center joint of the poselets window in the upper body visual feature space. Our poselet-regressor can also be regarded as the regressive version of the label-grid classifier [25], a visual grid classifier of the lower body joint location from the pelvis center with HOG-grid resolution. The center of alignment of the HOG features window in the label-grid classifier [25] is the pelvis center, while the alignment center of the poselet-regressor is the head center position in this paper.…”

“…Our poselet-regressor can also be regarded as the regressive version of the label-grid classifier [25], a visual grid classifier of the lower body joint location from the pelvis center with HOG-grid resolution. The center of alignment of the HOG features window in the label-grid classifier [25] is the pelvis center, while the alignment center of the poselet-regressor is the head center position in this paper.…”