SMAT focusses on developing intelligent sensing technologies to support and enforce traffic management of the SRT layer. This involves the development of robust and secure vision based sensing methods for collaborative computing and crowd sourcing. The underlying research in SMAT will lead to a better understanding of the advantage and limitations of utilizing real-time data to assist travellers in an Agile Transportation System.
Architecture-Aware Algorithms for Low-Cost Sensing
In order to provide for scalable vision based crowd sourcing for a dense urban public transport system, such as in Singapore, low-complexity sensing techniques are critical to justify mass volume adoption. Architecture-efficient algorithms are being developed to implement low-complexity and real-time vision-based traffic sensors. The robustness of such algorithms will be evaluated to ensure they withstand the varying weather conditions and complex conditions on a road segment.
Techniques for Vehicle Breakdown and Accident Detection
Vision based techniques are being devised to accurately establish traffic incidents and traffic violations. Computationally inexpensive computer vision algorithms are being developed to perform localized monitoring of current traffic condition. Vision based sensors will be used to detect traffic law violations (e.g. entering priority lanes, illegal parking in priority lanes, illegal turns, etc.). It is envisaged that mass-deployable real-time law enforcement solutions would ultimately ensure smooth traffic flow for SRTs.
Real-Time Scene Understanding for Congestion Management
Persistent congestions can be automatically identified through real-time scene understanding techniques. Crowd sourcing methods are being adopted by installing scene understanding sensors in vehicles to detect congestion. The relevant authorities can then take necessary actions to alleviate such unanticipated persistent congestion patterns that emerge over time.
Passive Sensing Techniques using Smartphone Location Data
The problem of inferring the mobility details of users from sequences of smartphone location samples involves considerable uncertainty, especially if the data is noisy and temporally sparse. Therefore, we intend to adopt a suitable probabilistic framework for solving this problem. To be able to process real-time data from a large number of users, the algorithms should have low latency and high computational efficiency. Therefore, are exploring optimization techniques to reduce the runtime and latency of the proposed methods. The developed solutions will be subjected to extensive evaluations using both real data as well as synthetic data obtained from traffic simulation environments. Robust statistical methods will be explored for aggregating information extracted from individual mobility traces in order to estimate the overall travel demand.
Secure Autonomous Traffic Security
A security-enabled sensory network protocol that satisfies real-time constraints is being developed using a virtual traffic simulation environment. The proposed network protocol will be deployed and benchmarked with off-the-shelf micro-controllers and FPGA development boards. Side-channel attack/tamper-resistant low-cost cryptographic accelerators for Internet of Things (IoT) communication will be developed such that the encryption/decryption modules are physically resistant to reasonably sophisticated side-channel attacks.