【摘要】 在复杂背景下的小型无人机红外目标检测是计算机视觉领域的挑战性课题.传统目标检测算法利用深度卷积神经网络提取无人机的静态外观特征并进行模式判别,但在复杂背景下且目标外观不清晰时的性能会显著下降.本文借鉴生物视网膜机制,通过视网膜大细胞通路模型提取无人机目标的时空运动信息,同时借助深度卷积神经网络获得基于静态表观特征的目标置信度图,进而将视网膜时空运动信息与深度卷积网络的目标置信度图进行融合获得目标检测结果.在Anti-UAV2020公开数据集上的评估结果表明,所提出算法的检测精确率达到86.90%,超过了业内主流的YOLO-v3算法.更多还原

【Abstract】 Video object detection of small unmanned aerial vehicles(UAVs) under complex background is a challenge in the field of computer vision. Many traditional methods are built on deep convolutional neural networks,mainly making use of static visual information while recognizing the object pattern. Nevertheless,their performance in complex background scenarios is yet far from the human being. Inspired by the retina mechanism,this paper designs a method that extracts spatio-temporal motion information by the magnocellular pathway model. At the same time,the object confidence map in terms of static visual features is obtained by the deep convolutional neural networks. Subsequently,the final object probability map is computed by fusing the static information and the spatio-temporal motion information. According to the evaluation results obtained on the anti-UAV2020 dataset,the proposed method yields a detection accuracy value of 86.90%,outperforming the widely used YOLO-V3 algorithm.更多还原