基于人工神经网络的多模型目标跟踪算法

Multi-Model Target Tracking Algorithm Based on Artificial Neural Network

摘要:

针对在目标跟踪中单模型跟踪算法难以应对目标运动形式的变化,而多模型跟踪算法存在结构固定、跟踪精度被非匹配模型削弱且模型切换缓慢的矛盾,文章提出了一种基于人工神经网络的多模型目标跟踪算法。通过分析目标几种基本运动模式的轨迹特点,归纳出目标运动轨迹的特征向量。利用训练好的 BP神经网络对滑窗里的轨迹段进行运动模型识别,按结果进行跟踪模型切换,达到使跟踪算法实时适应目标运动状态的目的。仿真结果证明了该算法的有效性,且与传统的多模型算法相比,具有结构更加简单、更强的灵活性和拓展性的特点。

Abstract:

For single-model tracking algorithm in target tracking, it is difficult to track maneuvering complex targets.Meanwhile, multi-model tracking algorithm has the disadvantages of fixed structure and model competition, which leads tothe decrease of tracking accuracy and model cut delay. A multi-model tracking algorithm based on artificial neural net?work wasproposed in this paper. By analyzing the trajectory law of the target three basic motion patterns, the feature vectorof the target trajectory was concluded. The BP neural network was used to identify the eigenvectors of the trajectory seg?ment in the sliding window, so as to switch the tracking model, so that the tracking algorithm could match the target motionstate in real time and improve the tracking accuracy. The simulation results proved the effectiveness of the proposed algo?rithm. Compared with the traditional multi-model algorithm, it has the advantages of simpler structure, more flexibility andmore expandability.