基于空间注意力机制的视觉多目标跟踪

Visual multi-object tracking based on spatial attention mechanism

在多目标跟踪方法中引入空间注意力机制,是克服遮挡与检测器漂移难点问题的有效手段.研究了一种空间注意力网络,并针对其在特征融合阶段易丢失不同通道空域结构信息的不足进行了改进,提出用交并比代替加权池化进行特征融合.实验结果表明,改进后模型有效提高了关联模型的精度,实现了性能良好的在线多目标跟踪.

Spatial attention mechanism is an efficient technique to deal with occlusions and detector drift in multi-object tracking.A spatial attention network is investigated aiming to improve its deficiency that ignores the spatial structure information existing in each channel.Specifically,Intersection over Union(IoU)is proposed to substitute weighted pooling as feature fusing strategy.The experimental results demonstrate that the proposed model can improve the accuracy in data association,achieving better tracking performance in an online manner.