基于动态特征融合的智能车应用检测分割技术  

Dynamic feature-fusion detection and segmentation technology for smart car applications

作　　者：舒鑫印 王萍 SHU Xin-yin;WANG Ping(College of Information Science and Technology,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学信息科学与技术学院,上海201620

出　　处：《计算机工程与设计》2020年第10期2838-2842,共5页Computer Engineering and Design

摘　　要：针对车辆行驶环境变化引起的视频场中运动目标大动态尺度变化、抖动、多场景下多目标检测分割等问题,提出一种基于视频的动态特征及边缘信息融合的运动车辆目标检测分割方法。利用特征提取所获视频场中光流场尺度信息与多方向边缘信息,通过逻辑“与”的二值信息融合技术对关联运动特征进行强化;设计自适应阈值三帧间差法预处理技术提取有效运动目标区域,提高算法检测速度,增强抗光照干扰能力。在多种场景下的实验结果表明,该方法有良好的检测性能指标,综合指标FM和检测速度分别达到90%和45帧/s。Aiming at dealing with the difficulties such as the large dynamic change of the scale,occlusion,jitter and multi-object tracking of the moving vehicles in the video field,an enhanced moving target detection and segmentation method based on video dynamic features and edge information fusion was proposed.Its characteristic was to use optical flow field scale information and multi-directional edge information in the rideo field obtained by feature estraction.The associated motion features were strengt-hened through the logical and binary information fusion technology.The pre-processing method based on the three-frame difference with an adaptive threshold was designed to refine the effective motion region,yielding a high detection speed as well as the anti-light interference capability.Experimental results in various scenarios show that the method has good detection performance index.In addition,the comprehensive evaluation FM and average processing speed are 90%and 45 frames/s respectively.

关 键 词：光流 边缘检测 动态特征融合 运动目标检测分割 动态阈值三帧间差 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]