Characteristic selection method based on tracking time prediction

The invention discloses a characteristic selection algorithm based on tracking time prediction, comprising the following steps: randomly selecting n-numbered characteristics of one-frame two-dimensional images to be tracked, obtaining the characteristic matching result and estimating the motion information of the two-dimensional images by utilizing the result; segmenting the two-dimensional images to obtain a group of image subregions which do not overlap each other, taking the central pixel of each subregion as the characteristic point, predicting the tracking time of the characteristic point by the forward iterative algorithm according to the motion information of the two-dimensional images and the current location of the characteristic point and taking the tracking time of the characteristic point as the predicted tracking time of any characteristic in the subregion of the characteristic point, namely the predicted tracking time of each subregion; and comparing the predicted tracking time of all the subregions and carrying out characteristic extraction in the subregion with the maximum predicted tracking time. The characteristic selection method provided by the invention can effectively reduce the uncertainty of self location estimation when being used for selecting characteristics from the environment by robots.