Tampere University: A content-aware pipeline ensures more efficient and faster video coding without compromising the quality
Tampere University is Finland’s second-largest university with seven faculties conducting scientific research in technology, health and society providing the highest education within these fields. Tampere University’s Ultra Video Group (UVG) is the leading academic video group in Finland and one of the largest in Europe. UVG is composed of over 20 experts with over 20 years of experience in conducting pioneering industry-driven research on visual media technologies.
In AISA, UVG is leveraging its visual media expertise by developing a content-aware vision pipeline for hybrid human-machine consumption in smart connected manufacturing. The primary focus is on advanced saliency-guided video coding techniques that seek to fuel efficient transmission of high-quality video over heterogeneous bandwidth-limited networks. Supporting reliable and real-time video streaming is of utmost importance to many industrial application domains, like remote machine control, production line maintenance, product quality monitoring, or worker guidance and safety.
Seamless integration of machine vision and video coding techniques
Traditional video coding tools are agnostic to video content being processed and thereby unable to detect Regions-of-Interest (ROIs) from it. This leads to suboptimal coding efficiency and visual quality in many industrial use cases.
The key innovation of the developed content-aware vision pipeline lies in the seamless integration of machine vision and video coding techniques so that the pipeline can dynamically adapt to the content being processed. Machine vision tools are used to extract ROI from visual data, and the ROI information is then applied to dynamically control video coding tools for optimized bit allocation.
One of the major findings is to use motion information from the video encoder to track the detected ROIs in real-time. This light feedback loop ensures that the encoding parameters are continuously updated in accordance with ROIs.
In AISA, the main outcome of the UVG’s research is a content-aware vision pipeline that improves coding efficiency and speed without compromising subjective visual quality in the salient areas of the video. Furthermore, the extracted ROI information can be used for automatic content annotation and tagging, enabling easy indexing and retrieval of relevant video content.
Assoc. Prof. Jarno Vanne, firstname.lastname@example.org
Dr. Alexandre Mercat, email@example.com