Correct detection of interferometric fringes is necessary to enable high resolution metric measurements using laser diode (LD) based self-mixing (SM) interferometry. For noisy, experimentally acquired SM signals, fringe detection is a challenge even when the LD remains mono-modal. This task becomes more complicated in case multiple laser modes undergo SM. In this paper, a novel SM fringe detection method is proposed which correctly processes multi-modal as well as typically encountered mono-modal SM signals. As SM fringes appear as local signal peaks in SM signals, so the proposed method uses peak detection to detect all-possible local peaks within the SM signal. Then, it successively refines its performance by removing false peaks (fringes) by analyzing corresponding SM fringe characteristics. In order to validate the proposed method for SM vibration sensing, several mono-, bi-, and tri-modal SM signals were acquired by using two different LDs normally emitting at 637 nm and 650 nm respectively. The proposed method is also tested on experimental SM signals corresponding to cases of variation in frequency as well as amplitude of remote target vibration. The proposed method has correctly detected 10786 fringes out of total 10844 fringes making up the data-set i.e. an accuracy of 99.46%. However, 55 false fringes were also detected (0.51%) while 58 true fringes were undetected (0.54%) by the proposed method.