基于模板修正和自适应Kalman预测的刚性目标跟踪

目前在远景刚性目标的跟踪中,由于长序列图像具有亮度动态范围大及背景噪声大的特点,当前模板的尺寸和位置往往不能有效代表目标,从而使目标的预测和相关搜索产生误差累积;另外,Kalman预测常因过程噪声与模型不匹配使其对机动目标跟踪适应性差.对上述问题进行研究提出了一种基于区域增长的模板修正方法,并对Kalman预测中过程噪声自适应的方法进行了仿真.结果表明,这种新的模板修正方法具有良好的尺寸及位置自适应能力和抗背景噪声能力,而且过程噪声的自适应也有效提高了Kalman预测的准确度,对目标跟踪具有指导作用.

A Rigid Object Tracking Algorithm Based on Template Correcting and Adaptive Kalman Prediction

In the process of tracking distant rigid object, because the illumination of image sequence has large dynamic range and image background is noisy, the size and position of current template often fail to represent, the object effectively, which causes the error in prediction and correlation - searching to accumulate progressively. Meanwhile, in Kalman prediction, if real noise is not matched with the model,Kalman filter cannot well adjust to maneuvering object. To solve these problems,a new algorithm of automatic correcting template based on region growing is brought forward,and an adaptive Kalman filter algorithm is used to self- adjust the process noise in prediction. The results show that this template size and position correcting ability improves tracking effect and adapts to complex background,and the adaptive Kalman algorithm increases the prediction accuracy.