Quasi-consensus of second-order leader-following multi-agent systems