Vehicle tracking is one of the essential technology for Smart City Systems. Existing vehicle tracking methods have achieved outstanding performance in simple traffic scenarios. However, these methods often fail to track specific traffic scenarios, such as vehicle deformation, lights, and occlusion between vehicles. In this paper, we propose a novel vehicle tracking method to solve the problem of difficult vehicle tracking and slow processing time in complex traffic scenarios. Our proposed framework incorporates strategies for short-term tracking and long-term matching. Short-term tracking is achieved through calculating the offset between consecutive frames, which has the characteristics of short-time consumption and high precision in traffic scenarios. Long-term tracking employs vehicle trajectory prediction and suspicious trajectory (ST) analysis in complex traffic scenarios. The tracking accuracy improved by matching the trajectory points and appearance features of continuous time series. The experimental results demonstrate that our method has superiority in Multi-target tracking (MOT) tasks in complex traffic scenarios.