摘要
针对东北虎图像收集效率低的问题,提出了基于无人机载摄像机收集东北虎图像的方案,提高了东北虎图像收集效率。针对多目标跟踪算法参数量大,准确率低的问题,提出了使用MobileV3和PW、Yolov4-Tiny的结合作为检测模型,使用SE-Resnet50作为重识别模型的DeepSORT的多目标跟踪算法方案。使目标检测模型参数量降低到3.384×10^(6),mAP达到了89.96%,浮点运算数为Yolov4-tiny的32.6%。多目标跟踪准确率达到51.8%,帧率达到8.54帧每秒。结果显示改进后多目标跟踪算法可以满足提高准确率和速度的需求。
In view of the problem of low Amur tiger image collection efficiency,in this paper,an Amur tiger image collection scheme based on UAV cameras was proposed,thus improving the Amur tiger image collection efficiency.Aiming at the problem of large number of parameters and low accuracy of multi-target tracking algorithm,a DeepSORT multi-target tracking algorithm scheme using MobileV3,PW and YOLOV4-TINY combination as a detection model and SE-Resnet50 as a re-recognition model was proposed,with the number of parameters in the target detection model reduced to 3.384×10^(6),mAP reaching 89.96%,floating point operand being 32.6%of YoloV4-TINY,the multi-target tracking accuracy reaching 51.8%and frame rate reaching 8.54 frames per second.The results show that the improved multi-target tracking algorithm could meet the requirements of improving accuracy and speed.
作者
徐其森
谢永华
XU Qi-sen;XIE Yong-hua(College of Mechanical and Electrical Engineering,Northeast Forestry University,Harbin Heilongjiang 150040,China)
出处
《林业机械与木工设备》
2022年第5期41-46,53,共7页
Forestry Machinery & Woodworking Equipment
关键词
无人机
目标检测
东北虎
多目标跟踪
UAV
target detection
Amur tiger
multi-target tracking
