Abstract: As the Interacting Multiple Model（IMM） algorithm on target tracking adopts the cooperative turn model as one of the models, the difficulty of realtime turn rate estimation and low precision leads to tracking accuracy decreasing, a IMM algorithm is proposed innovatively based on new adaptive turn model. The trajectory feature vector is constructed by using three methods of mapping turn rate, and the average turning rate of the trajectory segment in the time window is estimated in real time through the trained BP（Back Propagation） neural network, so that the accuracy of turning rate estimation is improved, and the tracking accuracy is improved. Simulation experiments show that the proposed IMM algorithm based on adaptive turning model has higher tracking accuracy than the IMM algorithm of traditional adaptive turning model when the turn rate is large, which has high scalability and development prospects.

Key words: target tracking, turn rate, interacting multiple model, Back Propagation（BP） neural network, feature vector

摘要： 针对采用协同转弯模型作为模型之一的交互式多模型（IMM）目标跟踪算法中，转弯率难以实时估计且精度不高造成跟踪精度降低的问题，创新性地提出了一种新的自适应转弯模型的IMM算法。利用三种映射转弯率的方法构造轨迹特征向量，通过训练好的BP神经网络实时估计时间窗内轨迹段的平均转弯率，提高转弯率的估计精度，从而提高跟踪精度。仿真实验表明，提出的自适应转弯模型的IMM算法较之传统的自适应转弯模型的IMM算法在转弯率较大时具有更加高的跟踪精度，并且具有较高扩展性和发展前景。

关键词: 目标跟踪, 转弯率, 交互式多模型, 反向传播（BP）神经网络, 特征向量