The ongoing large-scale deployment of 5G (3GPP New Radio) cellular systems makes it attractive to use their signals for “opportunistic” bistatic radar sensing, where the base stations serve as transmitters and user equipment (UEs) as receivers. Such 5G based radar sensing has the potential to complement and improve existing advanced driver assistance systems (ADAS) and future self-driving cars, e.g., in terms of increased sensing range. However, realizing a radar sensing system using strictly standards-compliant 5G-NR signals is a complicated task, mainly because of the signal inconsistency across time and frequency. This paper first analyzes the different reference signals (RSs) defined in the NR standard and which combinations thereof are suitable for vehicular radar applications. We then present several signal processing approaches to overcome difficulties arising from the standardized signals; in particular two-dimensional interpolation between non-regularly spaced locations in the time-frequency plane, where the classical Nyquist-Shannon sampling breaks down, as well as serial interference cancellation(SIC) for determination of the peaks in the delay-Doppler domain with better-than-Fourier resolution. We finally demonstrate the use of payload data for improving the radar estimates. Simulation results demonstrate the validity of our approach.