| CPC G06T 7/262 (2017.01) [G06T 2207/10016 (2013.01)] | 20 Claims |
|
1. A target tracking method, comprising acquiring N frames of image sequentially, wherein N is an integer greater than or equal to 2,
wherein the target tracking method further comprises:
performing target detection on an N-th frame of image by a target detection algorithm to generate a target area;
performing optical flow calculation on a target area in an (N−1)-th frame of image to generate an optical flow predicted target area in the N-th frame of image; and performing Kalman Filter prediction on the target area in the (N−1)-th frame of image to generate a Kalman Filter predicted target area in the N-th frame of image;
performing a first matching between the target area in the N-th frame of image and the optical flow predicted target area in the N-th frame of image;
updating a Kalman Filter model and performing an optical flow point resampling for a target area in the N-th frame of image which passes the first matching, in response to the target area in the N-th frame of image being matched with the optical flow predicted target area in the N-th frame of image;
performing a second matching between the target area in the N-th frame of image which does not pass the first matching and the Kalman Filter predicted target area in the N-th frame of image, in response to the target area in the N-th frame of image being not matched with the optical flow predicted target area in the N-th frame of image;
updating the Kalman Filter model and performing the optical flow point resampling for the target area in the N-th frame of image which passes the second matching, in response to the target area in the N-th frame of image which does not pass the first matching being matched with the Kalman Filter predicted target area in the N-th frame of image;
creating and initializing a Kalman motion model for the target area in the N-th frame of image which does not pass the first matching or the second matching, and performing optical flow point sampling, in response to the target area of the N-th frame in image that does not pass the first matching being not matched with the Kalman Filter predicted target area in the N-th frame of image;
performing prediction condition determination on the Kalman Filter predicted target area in the N-th frame of image which does not pass the second matching;
deleting the Kalman motion model in response to the determination being that the prediction condition is not met; and
updating the Kalman Filter model and performing the optical flow point resampling by using the optical flow predicted target area in the N-th frame of image corresponding to the Kalman Filter predicted target area in the N-th frame of image which does not pass the second matching, in response to the determination being that the prediction condition is met, wherein the Kalman Filter predicted target area in the N-th frame of image which does not pass the second matching and the corresponding optical flow predicted target area in the N-th frame of image correspond to a same target.
|