摘要
针对低能见度环境中人员的监测和保护问题,提出了一种基于多特征级联的红外行人检测方法,利用感兴趣区域长宽比特征和头部Haar特征组成初级分类器,改进的HOG-SVM完成最终行人识别。所提出的改进的HOG特征提取算法和自适应缩放因子获取算法,在保证检测率的基础上,有效地减少了帧间处理时间,针对目标被遮挡情况,提出了遮挡情况判断和局部特征识别功能,由此进一步提高了算法应用于复杂工况下的鲁棒性。实验表明:该检测方法能够达到91%的检测率,较现有算法性能得到提升,同时也满足了系统实时监测要求,适用于低能见度、粉尘的工况作业环境。
Aiming at the problem of personnel monitoring and protection in low visibility environment, an infrared pedestrian detection method based on multi feature association was proposed, the primary classifier was constructed by using the aspect ratio of interest region and the Haar feature of head, and the improved HOG-SVM was used to complete the final pedestrian recognition. An improved HOG feature extraction algorithm and an adaptive scaling factor acquisition algorithm were proposed, and the interframe time was effectively reduced on the basis of guaranteeing the detection accuracy. In view of the occlusion of the target, the occlusion detection and local feature recognition were proposed, which further improved the robustness of the detection system under complicated circumstances. The experimental results show the detection method can achieve the detection rate of 91%, which is better than the existing algorithms, and also meets the real-time monitoring requirements of the system. It is suitable for low visibility and dust working environment.
作者
刘峰
王思博
王向军
赵广伟
霍文甲
Liu Feng;Wang Sibo;Wang Xiangjun;Zhao GuangWei;Huo Wenjia(State Key Laboratory of Precision Measuring Technology and Instruments,Tianjin University,Tianjin 300072,China;MOEMS Education Ministry Key Laboratory,Tianjin University,Tianjin 300072,China)
出处
《红外与激光工程》
EI
CSCD
北大核心
2018年第6期127-134,共8页
Infrared and Laser Engineering
基金
国家自然科学基金(51575388)