Multi-model induced network for participatory-sensing-based classification tasks in intelligent and connected transportation systems

Classification is an emerging Internet of Things (IoT) application in intelligent and connected transportation systems (ICTS). To reduce the costs associated with classification tasks, the concept of participatory sensing has been integrated into ICTS. In this approach, for performing classification tasks, a crowd of participants reports multi-modal data collected via smart devices and manual marking. However, the multi-modal data thus collected are always incomplete. Therefore, integrating these multi-modal data for classification in ICTS is a challenge. In this paper, we propose a multi-model induced network (MMiN) framework for participatory-sensing-based classification tasks in ICTS. We first explore the relationships of the multi-modal data through graph modeling and hypergraph learning. Then, the derived relationships and the multi-modal data are used to train the MMiN. Finally, end-to-end features are derived by the MMiN as the classification results. To evaluate the effectiveness of the MMiN framework, experiments conducted on three visual datasets and one trajectory dataset are reported. The experimental results and comparisons demonstrate the effectiveness of the MMiN framework in exploiting incomplete multi-modal data.