基于可变形部件模型的渔船安全监控系统

摘要

The video-based human detection on fishing boat crew is a guarantee for a safe trip.To improve the accuracy of human detection with different gestures, multi-scale and occlusions in confined fishing boat cabins, a new monitoring system is proposed.It remodels this specific scene by changing weights of different part filters and optimizing scoring function of detection.Non Maximum Suppression methods are also optimized to solve the problems of duplicate or missed detection caused by occlusions.At last,it uses message transmission of the Beidou navigation system to monitor the safety of inshore fishing boats.It is verified by experiments that the system increases the accuracy of human detection and counting,and meets the needs of safety monitoring of inshore fishery.%基于视频的渔船驾驶舱人员检测是渔船安全行驶的保障.为解决渔船驾驶室小范围场景下人员的多姿态、多尺度、遮挡严重造成的检测精度不高的问题,系统针对场景重新建模,优化各组件权重和改变可变形部件模型的打分函数,并优化非极大抑制法,进一步缓解人员遮挡造成的多检和漏检问题,最后利用北斗卫星导航系统的报文传输功能实现对近海渔船的驾驶舱安全监控.实验结果表明,系统进一步增加了该场景下人员检测和人数统计的精度,满足了渔业安全监控的需要.