算法-IMM

trajectory-prediction程序的imm.cc中的以下代码的对应的算法原理在后面

void IMM_UKF::InputInteract() {if (std::isnan(model_pro_(0)) || std::isnan(model_pro_(1)) || std::isnan(model_pro_(2)))std::abort();if (model_pro_.sum() != 0)model_pro_ /= model_pro_.sum();c_.fill(0.0);// 遍历所有模型对，根据交互概率interact_pro_(i, j)和模型概率model_pro_(i)，计算每个模型j的混合概率c_(j)。// 这个混合概率表示在考虑模型间交互后，模型j被选中的概率。for (int j = 0; j < model_size; ++j) {model_X_[j] = imm_ukf_[j].Get_state();model_P_[j] = imm_ukf_[j].Get_covariance();for (int i = 0; i < model_size; ++i) {c_(j) += interact_pro_(i, j) * model_pro_(i);}}for (int j = 0; j < model_size; ++j) {X_hat_[j].fill(0.);P_hat_[j].fill(0.);for (int i = 0; i < model_size; ++i) {double u = ((interact_pro_(i, j) * model_pro_(i)) / c_(j));X_hat_[j] += u * model_X_[i];}for (int i = 0; i < model_size; ++i) {double u = (interact_pro_(i, j) * model_pro_(i)) / c_(j);P_hat_[j] += (u * (model_P_[i] + (model_X_[i] - X_hat_[j]) * (model_X_[i] - X_hat_[j]).transpose()));}}
}