摘要： 针对在大位移、弱纹理等情况下光流信息计算误差较大的问题,提出一种基于改进V1-MT前馈模型的光流计算方法。以视频序列作为输入,分别计算初级视皮层和中颞叶(MT)细胞的响应,分解MT细胞的响应得到光流信息。采用多尺度和由粗到精的方法,解决大位移情况下的光流计算问题,采用基于双边滤波的插值方法,融合邻域运动信息,以估计弱纹理区的光流信息。仿真结果表明,与其他基于生物模型的计算方法相比,该方法能更为精确地计算光流信息。

关键词: 光流, 运动信息, 大位移, 多尺度, 双边滤波

Abstract: In order to solve the problem of great computational error of optical flow information under large displacement and weak texture,an optical flow computational method based on an improved V1-MT feedforward model is proposed.By taking video sequences as inputs,the responses of cells in primary visual cortex and Middle Temporal(MT) are calculated.The optical flow information can be obtained by decomposing MT response.The multi-scale and coarse to fine methods are used to solve the optical flow computational problem under large displacement.The interpolation method based on bilateral filtering is adopted,so that neighborhood motion information can be fused and the the optical flow information in the weak texture region can be estimated.Simulation results show that this method can calculate optical flow information more accurately than other methods based on biological model.

Key words: optical flow, movement information, large displacement, multi-scale, bilateral filtering

中图分类号: 

• TP183