Text

theses

With the evolution of technology, vehicles are becoming increasingly connected and automated. They have evolved into rich sensing platforms with a plethora of diverse sensors, generating large amounts of real-time data including the location information, speed, acceleration, steering angle etc. With the recent advances of intelligent vehicle systems and Dedicated Short Range Communication (DSRC) devices, this data is broadcasted multiple times per second, so that each vehicle can be aware of nearby vehicles. However, such systems may fail to provide the location information if DSRC is not supported by one of the vehicles or in the case of a GPS unavailability and some of the shared data over DSRC or other networks could jeopardize user privacy. In this work, we focus on developing techniques that improve available location information and demonstrate the driver specificity of the shared data over such systems. GPS is widely used in critical infrastructures but is vulnerable to radio frequency (RF) interference. A common source of interference are commercial drivers that use GPS jammers to circumvent vehicle tracking systems. Existing mechanisms to detect and identify such interference emitting vehicles on roadways require a large number of specialized detectors or a manual observation process. To detect GPS jammers on roads, we designed a system that could detect any transmission at GPS L1 frequency. The key components of the system are monitoring stations (which are equipped with directional antennas and cameras) and mobile detectors (e.g., smartphones). Using an off-the-shelf software-defined radio (USRP) to emulate GPS jamming signals, we conducted a case study evaluation of our system with multiple trial drives on local highways in 2 US cities and found the monitoring stations effective. Through our experiments on a local highway with a vehicle transmitting interference in the 900MHz ISM band, we found that the vehicle identification rate of our mechanism is 65% for a single-point setup and 100% for a two-point setup. To study the privacy of the data shared among vehicles, we designed a system that can access a rich set of in-vehicle sensor data through a custom CAN bus interface and examined its driver specificity. We designed classifier features that allow distinguishing drivers based on a minimal set of sensor data. We evaluated the system with data from 480 real-world trips collected over 3 weeks from five university mail vans, with 24 drivers in a controlled experiment, and 103 trips with four drivers across two households. Our system could achieve 91% accuracy within the 20s after the driver enters the vehicle in the real world experiments. While the stream of rich sensor data can be communicated to and processed in a remote cloud, bandwidth and latency challenges encourage computation of this data on the vehicles themselves. With high computing powers and less power consumption, vehicles can sense the dynamic environment like no other platform. We propose to use harvesting vehicles as edge compute nodes, focusing on sensing and interpretation of traffic from live video streams. This work proposes effective fine-grained traffic volume estimation using in-vehicle dashboard mounted cameras. With the proposed system, we collect the footage of the traffic, detect vehicles using a real-time object detection method and estimate the lane of travel with the speed information for vehicles that are traveling in both directions. With such an information, not only the current positions of vehicles but also the estimated future positions of vehicles could be shown on a map. We conduct studies on different roads, our vehicle detection accuracy is over 75% even for highly occupied roads, and our speed estimation error is less than 12%. We could also estimate the lane of travel with over 80% accuracy.

ETD_8119

Ph.D.

Includes bibliographical references

by Gorkem Kar

doi:10.7282/T3SB48MK

ETD doctoral

The author owns the copyright to this work.