Vehicular Fog Computing (VFC) is an extension of fog computing in Intelligent Transportation Systems (ITS). It is an emerging computing model that leverages latency-aware and energy-aware application deployment in ITS. In this paper, we consider the problem of multi-hop computation offloading in a VFC network, where the client vehicles are connected to fog computing nodes by multi-hop LTE access points. Our scheme addresses three key aspects in a VFC architecture namely: (i) Optimal decision on local or remote task execution, (ii) Optimal fog node assignment, and (iii) Optimal path (multi-hop) selection for computation offloading. Considering the constraints on service latency, hop-limit, and computing capacity, the process of workload allocation across host vehicles, stationary and mobile fog nodes, and the cloud servers is formulated into a multi-objective, non-convex, and NP-hard Quadratic Integer Problem (QIP). Accordingly, an algorithm named Computation Offloading with Differential Evolution in VFC (CODE-V) is proposed. For each client task, CODE-V takes into account inter-fog cooperation, fog node acceptance probability, and the topological variations in the transportation fleets, towards optimal selection of a target fog node. We conduct extensive simulations on the real-world mobility traces of Shenzhen, China, to show that CODE-V reduces the average service latency and energy consumption by approximately 28% and 61%, respectively, compared to the state-of-the-art. Moreover, the CODE-V also gives better solution quality compared to standard $DE∕rand∕1∕bin$ algorithm and the solutions generated by a CPLEX solver.