Edge intelligence has boosted Internet of Vehicles (IoV) capabilities through collaborative sensing data, advancing smart traffic systems and autonomous driving technologies. However, several problems remain in the utilization of edge collaborative sensing data within the IoV: 1) Environmental constraints and limitations of sensors introduce noise; 2) Unreliable communication channels and sensor malfunctions result in data corruption during both acquisition and transmission. To address the aforementioned problems, this paper proposes an unsupervised denoising matrix completion (UDMC) method to overcome noise pollution and data corruption in edge collaborative sensing scenario within the IoV. Firstly, the UDMC method applies the variational autoencoder (VAE) to map complex noisy data into a latent space adhering to the additive white Gaussian noise (AWGN) assumption. Then, the UDMC method utilizing robust matrix completion for data recovery and denoising. Extensive experiments conducted on both synthetic and real-world datasets demonstrate that the proposed UDMC exhibits superior robustness compared to the baseline methods. Specifically, on the real datasets CC and NuScenes, the PSNR and SSIM achieved by UDMC are improved by at least 8.2% and 24.3%, respectively, compared to the baseline methods.