To reduce the noise interference and peak positioning error of the spectral system when the position of the object to be measured changes, based on the principle of Kalman filtering algorithm, a method is proposed to improve the measurement accuracy of the confocal system by taking the results of Voigt peak-finding and positioning as the observation error and making the optimal estimation. In this work, calibration experiments are carried out to determine the measurement range and accuracy of the spectral confocal system. Then, the denoising processing of spectral signals such as median filtering, Savitzky-Golay filtering, and fast Fourier transform filtering is compared in turn, and Gaussian, Lorentz, and Voigt fitting methods are used to find peaks. At the same time, the effects of peak extraction, Gaussian width, Lorentz width, and amplitude error on fitting accuracy in Voigt fitting are analyzed. The experimental results show that the measurement range of the system can reach 3 mm, and the radius of the central spot changes by nearly 1.79 times, The Kalman filtering algorithm can reduce the positioning error caused by noise in the system, and the system accuracy can be improved by 11 times to meet the requirements of high-precision measurement.