Innovative algorithm of the object detection in videos at the ISM 2019

Hannes Fassold, senior researcher at JOANNEUM RESEARCH - DIGITAL, gave a talk at the IEEE ISM conference about resource-efficient object detection by sharing backbone CNNs. This work (jointly done with Werner Bailer) addresses the problem of sharing a backbone CNN for different tasks, for example, to enable detection of additional classes when an already trained network is available. Furthermore, he presented a poster about realtime detection and tracking of objects, text and logos in conventional as well as 360° video. It combines a powerful Deep Learning based object detector (YoloV3) with high-quality optical flow methods.

Both works are very important for many application cases (like realtime video processing, autonomous drivng and edge computing), where it is crucial to have a lightweight network which can operate in realtime on the device.