Balance maintenance during human walking, as an energy efficient and obstacle adaptive motion, provides design guidelines and control strategies to develop biped robots. Synergy of muscle contractions, as the main power source for human voluntary motions, generates joint torques for the gait balance. However, the traditional electromyography approach is subjected to serious environmental disturbances and personal variations, and it is desired to develop a robust sensing method to monitor muscle motions. This paper proposes a sensing system based on muscle deformations to investigate muscle synergy for steady and unsteady human walking, revealing the contribution of gluteus maximus, rectus femoris and medial gastrocnemius to the lower limb joint motions in a gait cycle. The consistency and reliability of the muscle synergy analysis is demonstrated with nine tested subjects, and it is expected to uncover the balance mechanism of bipedal walking that is valuable to research of gait biomechanics, wearable robotics and humanoid robots.