Long-Term Evolution Vehicle-to-Everything (LTE-V2X) is a comprehensive communication solution tailored for vehicle-road collaboration to fulfill the diverse requirements of the Internet of Vehicles. To bolster the security performance of LTE-V2X, the hardware characteristics of transmitters, known as radio frequency (RF) fingerprints, can serve as unique identity IDs for LTE-V2X device authentication. In this paper, we present a scheme to extract RF fingerprints by recovered standard Demodulation Reference Signal (DMRS) sequences from received LTE-V2X signals. Building upon this, we propose a cepstrum domain cross-correlation method to enhance the fingerprint features and mitigate the impact of the channel on RF fingerprints. Subsequently, we select 10 On-Board Units (OBUs) to collect LTE signals in three scenarios (device direct connection scene, moving scene, and stationary scene) for experimentation. Results reveal that the recognition accuracy can achieve 94.3% at a signal-to-noise ratio (SNR) of 25dB and improves from 76.4% to 80.9% in the fingerprint cross-test of moving and stationary scenes after channel suppression.