The short answer is:
A poselet describes a particular part of the human pose under a given viewpoint.
The quote comes from this paper "Poselets: Body Part Detectors Trained Using 3D Human Pose Annotations" where I think the term may have been coined.
In the paper mentioned above, they are used as intermediate features in a computer vision pipeline for torso detection, keypoint detection, and people detection. The researchers train a large number of SVM models to identify specific poselets, which are then fed as features into a final top level model which performs the learning tasks. You could see this as feature engineering for a task-specific type of model stacking.