The monitoring of vital signs in a dynamic environment is challenging. This work demonstrates an approach to estimate the respiratory rate (RR) under real-driving conditions by using two accelerometers for signal recording and de-noising. One accelerometer was attached to the seatbelt for recording respiratory movements; another one was attached to the left side of the car seat for recording noise. The frequency components of the noise were used to suppress the noise hidden in the signal. The performance of the proposed approach is evaluated for three testers under three driving conditions, i.e., engine on, flat road and uneven road. The estimated RRs for three testers are 11.54 ± 2.28 breaths per minute (bpm), 15.57 ± 5.77 bpm, and 9.63 ± 4.58 bpm. The median estimated RR for three testers are 12.08 bpm, 18.26 bpm, and 7.76 bpm, where the manually counted reference RRs are 12 bpm, 18 bpm, and 7 bpm respectively. The average difference between estimated RRs and reference RRs is 0.71 bpm for the condition engine on, 3.36 bpm for flat road, and 4.58 bpm for uneven road. The results exhibit the ability of the proposed approach to estimate RR under real-driving conditions.