Dynamic traffic analysis and safe autonomous driving

A sensor system is being developed for flexible and weatherproof traffic analysis and for improving the safety of driver assistance systems as part of the KonSensData project of the Fraunhofer Cluster of Excellence Cognitive Internet Technologies CCIT. A mobile sensor system can collect traffic data at desired locations along the road, and can also compile and analyze it through AI. The results can then be used for tasks such as optimizing traffic flow at traffic lights, and can even be relayed to automated driving systems through V2X communication in order to enhance the vehicle's field of vision. 

© THI / Severin Mantel-Lehrer

Traffic optimization is a continuous process. Traffic data taken over different times and locations is necessary for improved decision-making. This must be collected using a privacy-compliant, weather-resistant and light-resistant method. The sensor system developed as part of Fraunhofer CCIT's KonSens project is undergoing further development in the follow-up KonSensData project to ensure flexible data acquisition through the mobile applicability of the sensors and to improve traffic analysis in terms of object classes and fields of view. Using this simplified data acquisition also unlocks further research and development potential regarding autonomous driving. Once obtained, the data improves the algorithms used for object detection and trajectory determination (both position and direction of movement) of road users in a 3D environment with the aim of further improving autonomous driving safety. 

Flexible data collection and linked analysis

There are many challenges when gathering traffic data. First of all, data must be collected and processed in accordance with data protection regulations. On a technical level, it is also important that the collection of data is minimally influenced by environmental conditions such as lighting conditions, weather, etc. The system must not require extensive adjustment in order to ensure flexible use. The project's objective is to develop a system that can overcome all of these challenges at once by the end of 2022. Data evaluation is essential here. An AI-based infrared image analysis that uses Convolutional Neural Networks will identify road users in the field of view. In the case of radar data analysis, various road users' point clouds are separated into clusters using AI algorithms. Then, micro-Doppler spectra signals are analyzed for movements so that road users can be classified. This heterogeneous data will be integrated with each other and converted into a list of objects in the traffic scenario. 

Three Fraunhofer Institutes

Three Fraunhofer institutes are working together on the KonSensData project. The Fraunhofer Institute for High Frequency Physics and Radar Techniques FHR is working on the hardware to develop detection and tracking algorithms based on the prototype created in the Horis and KonSens projects with the aim of reducing the data volume and processing movement data obtained from the micro-Doppler information. The Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS is providing algorithms that localize, classify and track road users by means of an infrared camera. The Fraunhofer Institute for Transportation and Infrastructure Systems IVI is merging the infrared camera, radar and accessories into one system, launching this sensor box in various German cities and supplying fusion algorithms for final 3D object detection and trajectory determination. On top of this, the Fraunhofer IVI is also developing the technology required to localize the sensor systems, which can then be used on the go.