基于仿生动态信号的航拍车辆检测算法

Vehicle Detection in Aerial Video Based on Dynamic Event Signal

现有主流航拍车辆检测算法均针对传统视频设计,但传统视频数据量巨大且存在信息冗余,这些缺点导致现有算法过于复杂且计算量大,运行效率低。提出基于仿生动态信号的航拍车辆检测算法,将整体检测系统拆分为候选目标定位模块与目标检测模块。候选目标定位模块以单个动态事件为输入,结合一定时空邻域内的局部动态事件群,快速定位候选目标。目标检测模块对传统ResNet结构进行改进,提出加入特征融合的M-ResNet以保护浅层特征。此外,制作了时长为25 s,含有295633872个动态事件的车辆事件数据集,通过该数据集验证该文提出算法可快速精准地检测出场景内的车辆。

In this paper,an aerial vehicle detection algorithm based on the dynamic event signal is proposed by splitting the detection system into a candidate target positioning module and a target detection module.The candidate target positioning module takes one single dynamic event as input,combining events within a certain spatial-temporal neighborhood to quickly locate candidate targets.Moreover,M-ResNet with feature fusion is proposed in target detection module to protect shallow features.Finally,this paper produces a vehicle event data set with a duration of 25 seconds and 295,633,872 dynamic events,and verifies that the algorithm proposed in this paper can quickly and accurately detect vehicles in the scene through this data set.