遮挡判别下自适应融合更新相关滤波跟踪

Adaptive Fusion and Updating of Correlated Filtering Tracking with Occlusion Discrimination

针对相关滤波跟踪方法在目标被遮挡和出现形变、运动等因素影响下跟踪效果较差的问题,提出遮挡判别下自适应融合更新的相关滤波跟踪方法.首先,利用分配权重的方式将目标的不同特征进行融合,从而得到初次融合特征,再进行第二次的特征自适应融合,并通过可信度策略得到最优融合特征,进而完成候选目标位置的确定;其次,在目标位置确定的基础上,根据候选位置的可信度判断是否启动遮挡检测机制,以确定目标最终位置;最后,根据预设更新阈值对模型进行自适应更新,保证跟踪器对目标的描述能力.在OTB100数据集进行实验,本文方法跟踪精准度和成功率分别为0.817和0.767,跟踪速度为40.1帧/s.实验表明本文提出方法能够在满足实时性前提下,可有效精准地跟踪目标.

As correlation filtering based tracking method has low tracking accuracy under the influence of occlusion and scale change,a tracking method is proposed by correlated filter tracking method for adaptive fusion update under occlusion discrimination.First,basic features are extracted and assign weights for fusion,and preliminary fusion features with different biases are obtained.Screening of the preliminary fusion features is performed for secondary adaptive fusion.The optimal fusion feature is selected as the tracking feature of the current frame target by using the credibility strategy to estimate target candidate position of the target.Secondly,the credibility of the candidate position is judged,and the occlusion detection mechanism is introduced to decide whether to start to determine the final position of the target.Finally,the model is updated adaptively according to the preset update threshold to ensure the description ability of the tracker to the target.To verify the proposed method,plenty of experiments on OTB100 were carried out.The tracking accuracy and success rate of the proposed method respectively are 0.817 and 0.767.The tracking speed is 40.1 frame/s.The experiment shows that the method proposed in this paper can effectively and accurately track the target on the premise of satisfying the real-time performance.