Robots, Autonomous Vehicles and Deep Learning

Recent technological improvements have led to an increase in performance and mobility of computation hardware. This development has enabled a high level of automation in traffic data acquisition and analysis. One of the most promising techniques in this field is image processing. While this technology shows huge potential due to being non-intrusive and having a high spatial coverage it often leads to thorough discussions on it’s ethical and social implications. Especially the issue of data privacy plays an important role in this discussion as image processing easily enables the operator to misuse the raw data to other non-goal-oriented purposes than traffic safety or management. In Germany, strict data protection measures restrict the use of the technology so new adapted analysis methods need to be developed and applied.

In this work the ethical implications of automatic image processing are presented and discussed especially focusing different applications of automatic number plate recognition (ANPR). This method based on Deep Learning has drawn huge interest in the field of intelligent traffic systems, as it can easily be applied in traffic management, traffic safety and law enforcement but can also be used for mass surveillance. In contrast, a new technique is presented which consists of detecting vehicle trajectories without gathering individual and privacy prone data such as vehicle license numbers and evaluating the dense trajectory data using specific indicators depending on the current application. The data required for analysis can be gathered from different sensor techniques such as CCTV cameras, camera equipped on unmanned aerial vehicles (UAVs) or thermal cameras by also using data fusion with laser scanner data. Based on different research projects, the application of this new technique is exemplified covering its use in traffic flow theory, traffic safety analysis, incident detection systems and optimization of traffic management systems.