Geo-temporal Online Analysis of Traffic Rule Violations
Examensarbete på kandidatnivå
Due to inattention and not complying with traffic regulations, human error accounts for roughly 94% of all traffic accidents. To counter this, the need to develop systems that can identify traffic rule violations and calculate the risk of collisions. The information reported can then be used to implement preventive measures. Modern vehicles are equipped with sensors and cameras thus making this possible, but it comes with the complication of not violating the privacy of individuals when gathering information. This project presents a prototype system comprised of three subsystems with the intention of reducing traffic accidents. The first two revolve around the detection of traffic violations with the use of real-time object detection and intention aware risk estimation. The purpose of the third subsystem is to detach personal information from the data gathered by the previously mentioned subsystems. This makes it possible to use the data to pinpoint problematic areas in a traffic environment. Evaluation of the system was performed in both a simulation environment and with analysis of video feeds from a lab environment. The results of the evaluation show that the prototype system developed in the project is sufficiently accurate to be further developed and implemented for use in real vehicles.
Traffic rule violations , Risk estimation , Privacy preservation , Computer vision , Deep learning neural net