主权项 |
1. A computer implemented method for tracking a target in a constrained environment in which motion of the target is constrained, the method comprising:
obtaining and storing video frames from a series of video frames on a computer readable storage device; calculating probability estimates for target locations in each frame for targets in the constrained environment; and determining candidate target locations in each frame, wherein the probability estimate for target locations in a video frame ‘t’ is calculated in accordance with:
P(xt)=p(xt|ot)p(xt|xt-1,xt-2, . . . x0)p(xt|ot)≈p(ot|xt)p(xt)where ot is a vector of the targets' observations for the video frame ‘t’, xt is a vector of targets' locations in the video frame ‘t’, xt-1 is a vector of the targets' locations in the frame ‘t−1’, P(xt) is a joint probability of targets' locations, p(xt|ot) is a probability of the target's locations given known observations, p(xt|xt-1, xt-2, . . . x0) is a probability of current targets' locations given known past target locations, p(ot|xt) is a probability of observations ot given known target locations xt and, p(xt) models a joint prior probability of the targets' locations, and at least one of:
a) wherein the target's locations are determined as a vector of locations that maximizes the probability for each location tar et is calculated in accordance with:x^t=argmaxxtP(xt)andwherein the target observation probability for each target is assumed independent thus yielding
p(ot|xt)=p(ot1|xt1)p(o2t|xt2) . . . p(otK×xtK)where superscript indices 1, 2, . . . K refer to tracked targets, and
b) the method further comprises simplification of the target propagation model and omitting interdependencies of propagation models among targets so that
p(xt|xt-1,xt-2, . . . x0)≈p(xt|xt-1,xt-2)p(xt−xt-1,xt-2)=p(xt1|xt-11,xt-21)p(xt2|xt-12,xt-22) . . . p(xtK|xt-1K,xt-2K)where superscript indices 1, 2, . . . K refer to tracked targets. |