10th EAI International Conference on Mobile Multimedia Communications

Research Article

Real Time Video Stitching by Exploring Temporal and Spatial Features

Download18 downloads
  • @INPROCEEDINGS{10.4108/eai.13-7-2017.2270086,
        author={Shaoen Wu and kelly Blair and Junhong Xu and Shangyue Zhu and Hanqing Guo and Kai Wang and Lei Chen},
        title={Real Time Video Stitching by Exploring Temporal and Spatial Features},
        proceedings={10th EAI International Conference on Mobile Multimedia Communications},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2017},
        month={12},
        keywords={interest point and salient region detections matching appearance and texture representations},
        doi={10.4108/eai.13-7-2017.2270086}
    }
    
  • Shaoen Wu
    kelly Blair
    Junhong Xu
    Shangyue Zhu
    Hanqing Guo
    Kai Wang
    Lei Chen
    Year: 2017
    Real Time Video Stitching by Exploring Temporal and Spatial Features
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.13-7-2017.2270086
Shaoen Wu1,*, kelly Blair1, Junhong Xu1, Shangyue Zhu1, Hanqing Guo1, Kai Wang2, Lei Chen2
  • 1: Ball State University
  • 2: Georgia Southern University
*Contact email: swu@bsu.edu

Abstract

Although image stitching has been investigated for years, realtime video stitching still lacks of e cient methods to meet the required frame rate for satisfactory human vision experience. This work pro- poses e cient video stitching solutions by exploiting both temporal and spatial features among video frames. As a result, the stitching speed is signi cantly improved with two techniques by exploiting: (1) the dimmension of distance (spatial) by focusing only on the region of frame overlap and (2) the dimmension of time (tempo- ral) by reusing homography information across multiple frames. Based on these two techniques, this paper presents three solutions to determine submiages for rapid stitching the video frames from side-by-side cameras. This work implements these solutions into a video stitcher. The evaluation over video streams shows that the proposed solutions can stitch the video at 6.5 frames per second (fps) in contrast to 1.5 fps in conventional imaging stitching approaches, which is over 400% improvement on stitching speed performance, but at the cost of a marginal drop in accuracy.