Evaluating the influence of packet loss on visual quality of perception for high bandwidth automotive networks

• A new automotive video dataset with mean opinion scores and saliency data.
• An automotive simulation testbed is used to generate packet loss impaired videos.
• A new objective no reference video quality metric is presented.
• The influence of packet loss events on visual attention is examined.
• Results of a subjective image quality test of videos with packet loss are presented.

This paper presents a new approach for modelling the perceptual quality of automotive video corrupted by packet loss. The model uses both temporal and spatial image statistics to distinguish between salient and non-salient packet loss events and hence derive an objective no-reference video quality metric. A new automotive specific video quality database is presented, consisting of 50 video sequences with associated human saliency data and mean opinion scores. Experiments carried out on this dataset investigate the influence of packet loss impairments on the visual attention of the viewer. The results suggest that packet loss impairments do not significantly alter visual attention. This is an important result, since it suggests that in the context of automotive vision, the visual attention of the driver is more strongly influenced by task related factors such as pedestrian detection, than by low-level sensory cues. The experiments further demonstrate that the proposed quality model correlates closely with human opinion for videos corrupted with network induced packet loss using a hybrid automotive testbed network simulation.