融合李群理论与特征子空间基的图像目标跟踪

针对复杂背景下目标跟踪窗口易受噪声干扰从而产生形变与漂移的问题,本文利用群空间中仿射群组受扰动后的形不变属性,将系统状态变量映射到李群空间进行处理,同时采用增量主元分析法(IPCA)算法实时学习并更新目标特征子空间数据.所提方法在利用粒子滤波算法采样粒子时,通过引入测量向量以提高权值计算的准确性.在基于Car11等4个测试集的实验中,结果优于IVT跟踪器,本文跟踪器窗口在噪声干扰下不会产生形变,跟踪成功率达到96%,结果优于IVT跟踪器.对比Kwon跟踪器,本文跟踪方法显著降低了算法复杂度,平均执行时间有效地控制在0.32s/帧.

Image object tracking on integrating lie group theory with characteristic subspace eigenbasis

To reduce the distortion and deformation of the object window in tracking objects with noises under complicated circumstance, we map the system state-variables to Lie group space for processing based on the affine-group invariability under disturbances. The incremental principal-component-analysis (IPCA) algorithm is employed for instant learning and updating characteristic subspace databases of the object. In sampling particles by using the particle filters, we introduce the measurement vector to improve the precision in weight-computation. In the testing of four standard video databases Car11, no deformation of tracker window caused by noises is found, and the successful tracking ratio reaches 96 percent. These results overtake those of the tracker IVT. When compared with tracker Kwon, the algorithm complexity is significantly lower and the average execution time is effectively kept within 0.32 s/frame.