Structural reliability analysis of complex systems is a challenging task; challenges are due to the small magnitude of the target probability, the complex shape of the failure domain that often includes multiple disjoint regions and the high dimensionality of the input random variables. In this work, we propose an enhanced version of sequential directional importance sampling (SDIS) to address these challenges. By magnifying the input variability, SDIS expresses the failure probability as the product of a group of integrals that are easy to estimate. The performance of SDIS depends on the choice of the initial magnification factor and the applied root-finding algorithm. To improve its flexibility, we reformulate the intermediate integrals in SDIS by simultaneously modifying the failure threshold and magnifying the input variability. Additionally, we propose an improved Markov Chain Monte Carlo (MCMC) algorithm to estimate the intermediate integrals. A Kriging-based active learning algorithm is further developed for finding the roots within a confidence interval on each directional sample. We demonstrate the performance of the proposed method with various complex benchmarks.