Abstract: Mobile edge computing in the Internet of Vehicles (IoV) can substantially reduce task transmission latency by providing proximate computing services, thereby significantly enhancing both the quality of service and the driving experience in intelligent transportation systems. However, the high mobility of vehicles and the dynamic variations of the environment make it difficult to accurately perceive the edge network state, while the heterogeneous resource demands across different vehicles further exacerbate the problem of excessive task offloading latency. To address these challenges, this work integrates mobile edge computing with digital twin (DT) technology to construct a DT-assisted vehicular edge computing network. In this network, physical entities are mapped to digital twins in real time, thereby precisely capturing the dynamics of both individual vehicles and the entire network, and providing a global view for offloading decisions without additional signaling exchange. On this basis, an offloading latency model that incorporates DT estimation deviations and task urgency is established. A vehicle selection algorithm based on Circle chaotic map is designed to identify candidate executors with high-quality computing capabilities, and an improved genetic algorithm integrating the sparrow search mechanism is further proposed to efficiently derive the optimal task offloading strategy. Experimental results demonstrate that the proposed scheme effectively reduces the total task execution time under various vehicle densities and task scales, possessing significant performance advantages over existing offloading solutions.