A method for quantifying classification confidence of obstructions applied to a perception mergence system of a vehicular computer in a vehicle. The method includes steps of: the vehicular computer receiving obstruction information of at least obstruction, image information corresponding to the obstruction information and vehicle body signals, and using a classifier to classify them; calculating a detection result of each range sensor to calculate a existence confidence; using the existence confidences and precision of the classifier to calculate a classification belief assignment of each range sensor corresponding to each obstruction; performing mergence calculation on the classification belief assignments to respectively quantify an obstruction classification confidence of all the range sensor corresponding to each obstruction; and performing a classification ineffectiveness filtering mechanism to exclude the obstruction whose obstruction classification confidence less than a predetermined value. The present invention quantifies the obstruction classification confidence to improve the classification precision.