Dynamic Vision Sensors (DVS) can capture dynamic changes in a scene due to their unique asynchronous response mechanism, but they do not respond to static targets. To address this issue, researchers have designed a polarization dynamic vision response system. By adding a polarizer in front of the DVS and rotating it, the brightness of partially polarized light changes, which can then be captured by the DVS. However, the system simultaneously captures two types of events: those caused by brightness changes from the rotating polarizer and those generated by actual motion. Failure to effectively distinguish between them leads to target misjudgment and significantly reduces the system's practicality. To solve this problem, this paper proposes a novel event stream separation method. First, a density-based spatiotemporal filtering algorithm is used for noise reduction and clustering. Then, a Time Surface is constructed to characterize the temporal distribution features of events. Finally, the Local Temporal Variation (LTV) is introduced to represent the degree of dispersion of event timestamps within spatiotemporal neighborhoods, achieving effective event separation. Experimental results demonstrate that the proposed method performs excellently in event stream denoising and separation tasks, providing new technical support for subsequent static target classification.