The objective of the Guardian Angel Project was to demonstrate that the mobile devices we carry and wear can provide effective safety services. This is particularly relevant where our devices contribute to dangers by causing distractions for drivers and pedestrians. This project, therefore, pursued the vision of a system that offsets such unsafe use by continually sensing our activities and surroundings, identifying potentially dangerous situations, and mitigating them through appropriate interventions. Specifically, the project focused on the following goals:
This project studied how wrist-mounted inertial sensors such as those in smart watches and fitness trackers, can track steering wheel usage and inputs. Identifying steering wheel usage helps mobile device detect driving and reduce distractions. Tracking steering wheel turning angles can improve vehicle motion tracking by mobile devices and help identify unsafe driving. The approach relies on motion features that allow distinguishing steering from other confounding hand movements. Once steering wheel usage is detected, it also use wrist rotation measurements to infer steering wheel turning angles. Our preliminary experiments show that the technique is 98.9% accurate in detecting driving and can estimate turning angles with average error within two degrees.
Toward Detection of Unsafe Driving with Wearables.
Luyang Liu, Cagdas Karatas, Hongyu Li, Sheng Tan, Marco Gruteser, Jie Yang, Yingying Chen, Richard P. Martin.
The 2015 workshop on Wearable Systems and Applications(held with MobiSys) . [pdf]
Determining Driver Phone Use by Exploiting Smartphone Integrated Sensors.
Yan Wang, Yingying Chen, Jie Yang, Marco Gruteser, Richard Martin, Hongbo Liu, Luyang Liu, Cagdas Karatas.
IEEE Transactions on Mobile Computing (TMC), 2015.
With the increasing popularity of wearable devices, the project also explored the potential to use wearables for steering and driver tracking. Such a capability would enable novel mobile safety applications without relying on sensors in the vehicle, thus making it more portable and accessible. In particular, the research investigated how wrist-mounted inertial sensors, such as those in smart watches and fitness trackers, can track steering wheel usage and angle. Tracking steering wheel usage and turning angle provided additional, fundamental techniques to improve driving detection, enhanced vehicle motion tracking by mobile devices and helped identify unsafe driving.
Leveraging Wearables for Steering and Driver Tracking.
Cagdas Karatas, Luyang Liu, Hongyu Li, Jian Liu, Yan Wang, Sheng Tan, Jie Yang, Yingying Chen, Marco Gruteser, Richard Martin.
The 2016 IEEE International Conference on Computer Communications . [pdf]
We also investigated using wearables to detect road crossings when a person is walking. Using the positioning information from only a smartphone, we were able to detect 85% of all pedestrian crossing events in suburban environments. Owing to degraded GPS performance in urban environments, we also used the inertial sensors on a foot-mounted unit to identify when a pedestrian is entering the street. In densely populated urban environments, we achieved a 90 percent detection rate with a less than 1 percent false positive rate.
LookUp: Enabling Pedestrian Safety Services via Shoe Sensing.
Shubham Jain, Carlo Borgiattino, Yanzhi Ren, Marco Gruteser, Yingying Chen, Carla-Fabiana Chiasserini
in Proceedings of the 13th International Conference on Mobile Systems, Applications, and Services (MobiSys), 2015 . [pdf][AT&T Connected Intersections Challenge Award][Conference Talk]
Recognizing Textures with Mobile Cameras for Pedestrian Safety Applications.
Shubham Jain, Marco Gruteser.
Preprint on arxiv.org [pdf]
Recognizing that automated driving promises some of the largest traffic safety gains, the project then branched out to harness mobile sensing devices to collect data about unusual driving scenarios. Most existing efforts to collect driving data build on a small fleet of tens of highly instrumented vehicles that are continuously operated with test drivers. In terms of miles recorded, it is challenging to accumulate a sufficiently large dataset with this approach. We took an alternate approach where a large set of minimally instrumented vehicles collect the training data needed machine learning approaches for self-driving and safety applications. The major advantage of our approach is that it provides a minimum-effort solution for self-driving companies and researchers, who want to collect large datasets of ready-to-use driving data in the real world without concerns of different vehicle types and driving behaviors. We evaluated the accuracy of collected internal and external data using over 140 real-driving trips collected in a 3-month time period. Results show that our approach accurately estimated the steering wheel angle with a 0.69 degree median error, and derived the vehicle speed with 0.65 km/h deviation. The approach was also able to determine wet vs. dry road conditions with 95 percent accuracy by capturing a small number of brakes.
BigRoad: Scaling Road Data Acquisition for Dependable Self-Driving.
Luyang Liu*, Hongyu Li*, Jian Liu, Cagdas Karatas, Yan Wang, Marco Gruteser, Yingying Chen, Richard Martin. (* indicates co-primary authors)
In Proceedings of the 15th ACM International Conference on Mobile Systems, Applications, and Services (MobiSys), 2017 . [pdf][Conference Talk]