摘要
为了解决孪生网络目标跟踪算法中互相关操作引入过多的背景干扰容易导致跟踪漂移的问题,提出了一种包含高精度相似度匹配且基于像素点回归的孪生网络目标跟踪算法。通过对Transformer的改进,设计出一种高精度的相似度匹配模块用于取代互相关系列操作,用以减少目标的背景干扰,并提出一种基于像素点回归的边界框预测策略用于取代锚框机制,用以提升算法的泛化性能。通过在目标跟踪数据集上的测试结果表明,该算法有效地提升了在包含背景杂乱、遮挡等挑战的复杂场景下的跟踪效果,并且可以实现实时跟踪。
In order to solve the problem that the cross-correlation operation in the siamese network object tracking algorithm introduces excessive background interference and easily leads to tracking drift,a siamese network object tracking algorithm including high-precision similarity matching and pixel-wise regression is proposed.Through the improvement of Transformer,a high-precision similarity matching module is designed to replace the series of crosscorrelation operations to reduce the background interference of the object,and a bounding box prediction strategy based on pixel-wise regression is proposed to replace the anchor mechanism,so as to improve the generalization performance of the algorithm.The test results on the object tracking datasets show that the algorithm effectively improves the tracking effect in complex scenes that include background clutter,occlusion and other challenges,and can achieve realtime tracking.
作者
刘庆玲
余晗华
LIU Qingling;YU Hanhua(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)
出处
《应用科技》
CAS
2022年第5期24-31,共8页
Applied Science and Technology