Skip to main content

Within the framework of the BMWi-funded joint project ifuse, algorithms and architectures for the fusion of sensor raw data at a low level of abstraction are being investigated. Compared to previous fusion processes at object list level, sensor data fusion at raw data level allows for a more robust classification of objects and detection of the vehicle environment, even if individual sensors are affected by environmental influences. Sensor data fusion at raw data level is based on signals from active and passive vehicle sensors (e.g. LIDAR, RADAR, camera, ultrasound), which after minimal preprocessing are related to a common coordinate system and located in an environmental model. Subsequent complex algorithms for the detection and classification of objects then process 3D points of the environmental model, which, in addition to the localisation information, contain further data, for example on the degree of confidence and environmental influences.

At the Department of Architectures and Systems of the Institute of Microelectronic Systems, data structures, algorithms and architectures for efficient sensor data fusion are researched. Based on an optimized 3D data structure, methods for open space and object detection and classification after fusion of the raw data are conceptualized and implemented. The results of the project will be demonstrated at the end of the project period in a prototypical vehicle of the TU Braunschweig at an urban intersection, which will be driven over autonomously.