The non-contact respiration sensing technology based on RF holds significant potential for monitoring human life states. In real-world scenarios, targets not only undergo their own motion but are also surrounded by moving objects. The signals reflected by these motions will nonlinearly overlap with respiratory reflex signals. Current solutions predominantly rely on the steady-state or transient characteristics of subject motion to separate signals with nonlinear superposition. However, they fail to effectively handle the random signals reflected by moving objects in the surroundings. To tackle this issue, we propose Pleno-Sense, which utilizes an adaptive switching algorithm to achieve robust respiration monitoring across diverse motion scenarios. Initially, the superposition characteristics of received signals in diverse motion scenarios are analyzed, highlighting the nonlinear superposition between signals reflected from various movements. Subsequently, a switching linear dynamic system is devised, allowing for the adjustment of model parameters based on different states or conditions. This facilitates better adaptation to the diverse statistical characteristics of signals. Finally, an extensive evaluation of the model is conducted using a dataset spanning 50 h. Experimental results show that Pleno-Sense accurately detects the target’s respiration rate across diverse motion scenarios, with an average error of 0.47 breaths per minute.