Integrating Dashcam Views through Inter-Video Mapping
Mu-Hsuan Chen1
Yan-Chiu Chen1
1National Taiwan University
2 University of California-Davis
abstract
In this paper, an inter-video mapping approach is proposed to integrate video footages from two dashcams installed on a preceding and its following vehicle to provide the illusion that the driver of the following vehicle can see-through the preceding one. The key challenge is to adapt the perspectives of the two videos based on a small number of common features since a large portion of the common region in the video captured by the following vehicle is occluded by the preceding one. Inspired by the observation that images with the most similar viewpoints yield dense and high-quality matches, the proposed inter-video mapping estimates spatially-varying motions across the two videos utilizing images of very similar contents. Specifically, we estimate frame-to-frame motions of each two consecutive images and incrementally add new views to form long-range motion representation. On the other hand, we dynamically infer spatial-varying motions across the two videos that propagated from local feature correspondences to trajectories. In this way, the observed perspective discrepancy between the two videos can be well approximated by our motion estimation. Since the inter-video mapping is established upon frame-to-frame motion, the correspondences can be updated incrementally with low cost, making it suitable for online applications. Experiments with real-world challenging videos demonstrate state-of-the-art reconstruction quality of our approach.
technical paper
presentation slide

Moving Cameras

Left: reference video (front vehicle), middle: target video (back vehicle), right: our result

HIGHWAY
BRIDGE
CITYROAD
UNDERPASS
HIGHWAY4
HIGHWAY5
HIGHWAY2
TUNNEL
TUNNEL2
STREET
CITYROAD2
HIGHWAY3
AELLEY

Different date rate

75kbps
200kbps
400kbps
1400kbps

Different frame rate


10 FPS
15 FPS

Different bridge selection approach


GPS
Quad
Ours
Ours (CSFR)

Network analysis


As the number of neighboring vehicles which interfere the information exchanges between vehicles is not constant, we also investigate the interference that individual vehicles experience under different traffic flows. We run packet-level simulation using NS-3 network simulator. The target pair of vehicles is surrounded by X pairs of neighbors, where X=0,...,8 is a simulation parameter. Results are obtained based on transmission of 100000 packets, where packet size is 1000 bytes. The simulation result is shown in Figure. In brief, with a minimum data rate requirement of about 100 kbps, the see-through application can work satisfactorily (without incurring noticeable packet loss) when the number of vehicle pairs that are concurrently running individual see-through applications is no more than 9.
bibtex

@article{CCCCC:2016,
     author = {Hsin-I Chen and Yi-Ling Chen and Mu-Hsuan Chen and Yan-Chiu Chen and Bing-Yu Chen},
     title = {Integrating Dashcam Views through Inter-Video Mapping},
     booktitle = {},
     year = {},
     pages = {},
}