基于深度学习的尺度自适应海面目标跟踪算法

Scale Adaptive Sea Surface Target Tracking Algorithm Based on Deep Learning

相比于普通场景的目标跟踪,无人艇海面目标跟踪具有目标尺度变化大、目标抖动剧烈和视角变化大等独特挑战。针对此,文中提出了基于深度学习的尺度自适应海面目标跟踪算法,以样本中心点是否落在真实目标框内对样本进行分类,直接回归中心点到目标框上下左右的距离预测目标框的位置和尺度。同时,建立了海面目标跟踪算法评估平台,以验证所提算法的有效性。试验结果表明,文中算法相比基于锚框的算法跟踪位置精度提升了4.8%,成功率提升了11.49%,有效解决了目标尺度自适应问题。

Compared with target tracking in common scenes,sea surface target tracking presents unique challenges such as changes in the target scale and perspective as well as intense dithering of targets.Accordingly,a scale-adaptive sea surface target tracking algorithm based on deep learning is proposed.The algorithm classifies samples according to whether the central point of the sample falls to the ground truth and then regresses the distances from the target location to the four sides of the bounding box to predict the position and scale of the target.An evaluation platform for the sea surface target tracking algorithm is also established to verify the effectiveness of the proposed algorithm.Experimental results show that compared with the anchor-based algorithm,the tracking accuracy of the proposed algorithm is improved by 4.8%and its success rate is improved by 11.49%,thus effectively solving the problem of target scale adaptation.