Robust Volumetric Reconstruction from Noisy Multi-view Foreground Occupancy Masks
We recover the shape of a moving object, represented by the occupancy of grid voxels in the physical space, from the consistency of its foreground occupancy in multiple camera views. To deal with the noise due to loose temporal synchronization, lighting variation and calibration errors, we integrate the foreground occupancy of each grid voxel over its local neighborhood. Especially, we design a way to decouple each integral direction, so that the accumulation along this direction can be efficiently computed by integrating pixels along a row/column in the transformed foreground masks, which also enables the access to the intermediate physical positions between two neighboring voxels. We further accelerate this integration by using the technique of integral images. The performance of the proposed method has been investigated experimentally.
Fan Chen Christophe De Vleeschouwer
Japan Advanced Institute of Science and Technology, Nomi, Ishikawa, Japan Universit catholique de Louvain, Louvain-la-neuve, Belgium
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-4
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）