Energy Harvesting Wireless Body Area Networks (EH-WBANs) where sensor nodes can harvest energy from their ambient environment are of great potential for low-power medical monitoring systems. The availability of energy in EH-WBANs, however, is challenged by uncertainty of harvested energy sources and limited charging efficiency, and thus the energy saving is still of great importance. To improve energy efficiency, this paper studies the sleep scheduling problem in EH-WBANs, which is of fundamental importance for network lifetime and connectivity while being not systematically addressed. Technically, this problem is proved to be NP-Complete, and thus needs non-trivial efforts to deal with. To that end, we present a series of approximation algorithms with proven performance from perspective of constructing minimum dominating set (DS), where energy saving and energy harvesting techniques are combined to prolong the network lifetime while guaranteeing real-time requirement of EH-WBANs. Specifically, we first propose a centralized algorithm that can construct DS with minimum size and discover the maximum number of DSs. We then design two distributed algorithms independent of prior global knowledge to improve network scalability, namely EEU and Improved EEU respectively. Theoretical analysis and extensive simulations are conducted to confirm the superiority of the proposed sleep scheduling algorithms.