In this paper, a novel approach towards generation and implementation of local reference trajectory points is presented. The need arises because Autonomous Vehicles (AV) are intensively dependent on the absolute self-localization with respect to a global map for their precise motion. However, in sev- eral practical scenarios (such as heavy rain on a highway), localization with low covariance is nearly impossible. As a result, we propose an alternate approach to generate a reference trajectory using local cues such as curbs or lane markers, which can then easily be tracked by the AV using a suitable control algorithm. The key idea is to project the curb/lane points about a Frenet frame along the AV longitudinal direction. Further, minimization of computational load and implementation of a temporal filter for tra- jectory smoothing over time are also examined. Finally, the proposed theories are verified in experiments to provide smooth reference trajectory points.