“Communication bandwidth and energy consumption requirements are important aspects to consider and fine-tune in a productive crowdsensing application,” Cherian said. “These requirements vary based on application, the level of accuracy required, the amount of data to be transmitted, and the rate (how frequently) at which it has to be transmitted to achieve a certain level of quality of service.” Cherian said large-scale crowdsensing applications and services could be delivered with today's level of connectivity, but newer technologies like 5G could only help.
When discussing vehicles, the natural question is: In a world of connected vehicles, could smartphones be taken out of the equation entirely in favor of data from in-vehicle sensors? “Vehicle sensor data are more reliable and they can be obtained using different methods. But this will demand some level of in-car instrumentation,” Cherian explained. “For example, the OBD2 [on-board diagnostics] connector that is installed in most cars manufactured in the last two decades offers some basic information about the vehicle that can be read wirelessly over Bluetooth or over a USB cable. More advanced vehicle information can be gathered using open connectivity protocols (such as OpenXC on Ford vehicles) or directly from the CAN bus (proprietary to each vehicle manufacturer) or using connected sensing devices including cameras.”
Cherian said that any additional sensor-rich hardware, if installed, can yield a wealth of information about the vehicle and/or its environment, which would in turn lead to more reliable services. “However, all of these methods require instrumenting the car, and may not be attractive enough to motivate a large crowd of users who may have their own concerns or reservations,” he said.
Cherian added, “It is a different story, though, if the local governments can also encourage or enforce the installation of connected on-board hardware for specialized purposes, such as parking/toll-collection or dynamic road pricing. For example, GPS-based ERP2 [Electronic Road Pricing] boxes are due to be rolled out across vehicles here in Singapore in the next few years. If they are crowdsensed, they can offer solutions to many other problems. On the other hand, many of the current proposals for vehicular mobile crowdsensing rely on infrastructure such as GPS and/or Wi-Fi. Unfortunately, they either do not work indoors or are unavailable to indoor parking garages—especially when underground.”
Cherian said research like ParkGauge, which proposes an infrastructure-free solution, becomes more relevant in this context. While the accuracy offered by mobile devices may be slightly inferior, it still offers a practically acceptable degree of accuracy. This is also where the use of machine learning plays a role, allowing for the development of crowdsensing systems that improve over time.
Learning from the Crowd
“It all depends on the quantity and, more importantly, the quality of data available for learning,” Cherian explained. In our experiments, we have made use of a hierarchical combination of ‘supervised’ machine learning methods. This means that we have utilized ground truth (in terms of labels that were manually provided by experienced human users or occupancy data that has been precisely obtained by infrastructure-based methods) so that the learning algorithm was able to learn patterns from it in a reliable manner.”
The ParkGauge researchers use a statistical analysis approach called a Hidden Markov Model that allows the algorithm to work with states that are not immediately observable. In other words, you don't have to actually see every vehicle to predict where it is or what it is doing.
“To detect the driving contexts (e.g., ‘driving’ or ‘parked’) from a smartphone placed in a vehicle, we use a Hidden Markov Model that can model the temporal evolution of contexts as a function of driving states (such as accelerating, braking, turning, or walking). These driving states can be easily detected from smartphone inertial sensor data, such as an accelerometer or gyroscope (and barometer, if available) using a decision tree-learning method, such as Random Forest classifier,” Cherian explained.
“Once we have the driving contexts and their time stamps, we can compute some temporal characteristics of parking garages and feed them into nonlinear regression methods to estimate the occupancy of the parking garage,” he said. “A primary reason why this works remarkably well is actually quite intuitive. The occupancy of a large parking garage is inversely correlated to the time taken to park there and the time spent in queuing or looking for an available space upon arrival. Combined with additional features or characteristics to learn from, we can obtain quite useful accuracy levels. Furthermore, as we acquire more data, we are able to progressively make better predictions on real-time parking occupancy for arriving drivers. We are also able to identify and report trends that are useful to businesses and parking operators alike.”
Following their 2016 study, the NTU Singapore researchers were hoping to test a large-scale deployment of ParkGauge, but Cherian said this hasn't come about due to internal factors and operational limitations. However, he said his team has explored methods to enhance the scalability of its crowdsensing solution while reducing the need to do a major data collection for deployments at new, unseen parking garages. All of this work is currently undergoing peer review, according to Cherian.
The team also recently published new research around another prototype crowdsensing system called ParkLoc, which is aimed at another parking-related pain point. “We have worked on yet another aspect of using infrastructure-free mobile sensing for indoor parking garages regarding the localization of a parked vehicle indoors,” Cherian said. “This work in particular is motivated by the often embarrassing problem of forgetting where we parked.”
Chris Wiltz is a Senior Editor at Design News covering emerging technologies including AI, VR/AR, and robotics.
|Today's Insights. Tomorrow's Technologies
ESC returns to Minneapolis, Oct. 31-Nov. 1, 2018, with a fresh, in-depth, two-day educational program designed specifically for the needs of today's embedded systems professionals. With four comprehensive tracks, new technical tutorials, and a host of top engineering talent on stage, you'll get the specialized training you need to create competitive embedded products. Get hands-on in the classroom and speak directly to the engineers and developers who can help you work faster, cheaper, and smarter. Click here to register today!