A method to detect people in multiple locations using a multiple-input and multiple-output (MIMO) frequency-modulated continuous wave (FMCW) radar system is proposed. A 2-D range-angle map is estimated by digital beamforming with virtual receiver arrays obtained via the operation of time-division multiplexing (TDM) MIMO radar. Phase variations due to human vital signs and movements allow the identification of people from stationary clutter in the range-angle map. To improve the sensitivity of phase-based human detection, a curve-length (CL) method that estimates the length of the I/Q signal trajectory is proposed. The proposed method can overcome the codomain limitation of a conventional method through its use of the standard deviation of phase variations. Measurements show that the method improves the signal-to-clutter ratio (SCR) by 10 dB compared with that when using the conventional approach.