Road Condition Detection using Commodity Smartphone Sensors Aided with Vehicular Data
Examensarbete för masterexamen
Computer systems and networks (MPCSN), MSc
Vehicles have become increasingly technologically advanced and have gone from being all mechanical to almost all computerized. Leveraging the digital vehicle and the many different sensors available, advanced vehicular systems have emerged that can intervene in critical situations, before the driver has a chance to react. The technical advancement realized, is however not limited to vehicles, as similar technological betterment can be observed in smartphones. Smartphones have emerged from being basic devices into an advanced platform housing different communication channels, capable computational hardware and access to a diverse set of sensors to interact with the surroundings. In this thesis, the problem of detecting a diverse type of road conditions is studied using data fusion between the available sensors, both from the smartphone and the vehicle. In the proposed system, the smartphone is the designated computational platform in addition of providing its build-in sensors to the detection algorithms. Positioning by the means of Global Positioning System (GPS) is used to tag detected events and by leveraging the accelerometer, vehicular shock and vibration information is inferred, all from the smartphone sensors. This sensory information is then complemented by the additionally available sensors from the vehicle such as speed, throttle position, motor revolutions per minute (rpm) and motor load. The main challenge faced comprises of fusing the data from the two disjoint sensory platforms together, to be able to detect road conditions. In addition, different sensor characteristics, such as sampling rate, have to be taken into consideration, as well as the limited computational capacity offered by a smartphone. Furthermore, battery consumption also has to be minimized for the proposed system to be a viable solution. Also, unlike some other related work which uses expensive calculations to perform the detection, a lighter approach is leveraged, without any heavy operations applied. The conclusions that can be drawn is that combining smartphone data with the vehicular sensors indeed helps in road condition detection. Not only does it help in determining the ground truth, which is a non-trivial problem to solve, but also to be able to detect different road conditions as well as to help reject falsepositives. Moreover, the developed algorithms presented are signaling the simplicity of the approach taken by means of leveraging the available sensors, ensuing their applicability as real-time algorithms to be executed on commodity smartphones.
Data- och informationsvetenskap , Computer and Information Science