Figure 1.

Illustration and example of the different components and their relationship. Left: Section 2.1: Selecting different pre-trained human pose estimation convolutional neural networks (HPE-CNNs) and performing inference, results in CNN detections. Section 2.2: Manually creating a labeled squash dataset by annotating selected videos, results in (ground truth) dataset labels. Section 2.3: Evaluating detections regarding labels using our evaluation procedure. Section 2.4: Utilizing detections and labels for heatmap visualizations. Right: Example detections (circles) of a CNN-based connected human body pose estimator overlaid on the processed video frame.