The main promise of intelligent transportation systems (ITS) is that leveraging the information sensed by millions of vehicles will increase the quality of the user's experience. However, the unpredictable nature of road events, combined with a projected network overload, calls for a careful optimization of the vehicles' data transfers, taking into account spatio-temporal, safety and value constraints. In this article, we provide a methodical solution to optimize vehicle-to-cloud (V2C) data transfers, based on a series of steps. First, we show that this optimization problem can be modeled as a soft real-time scheduling problem. Second, we provide an extension of a classical algorithm for the generation of workloads, by increasing its coverage with regards to our use-case representation. Third, we estimate the bounds of an optimal clairvoyant algorithm in order to have a baseline for a fair comparison of existing scheduling algorithms. The results show that, within all these algorithms, one clearly outperforms the others regardless of the load rate. Interestingly, its performance gain increases when overload grows, and it can be implemented very efficiently, which makes it highly suitable for embedded systems.