基于监控视频流的手持探针探测位置检测算法的设计

Algorithm Design for Position Detection of Hand-Held Probe based on Video Surveillance

摘要:

针对监控视频流,开展有效信息检测存在计算量大、耗时长等问题。通过构建视频图像数据集,训练人工手持探针模型；采用KNN算法分析前后帧的监控视频流,实现前、背景分离；利用人工手持探针模型实时提取监控视频中的探针,获得手持探针的前景图像；建立探针位置探测模型,并对比理论应检的真实位置,判断探针检测的有效性；实检结果表明,设计的手持探针探测位置检测算法的平均准确率约93.26%,召回率约81.11%,F1值约86.76%,检测速约9.66fps/s,能够实现工程监控视频流中手持探针的有效性检测。

Abstract:

Aiming at the problem of large calculation and long time consumption for the validity detection of video surveillance in factories. The dataset is established based on the factory surveillance video, and the model of hand-held probe is trained; K-Nearest-Neighbors (KNN) algorithm is applied to analyze the continuous frames of video and separate the foreground and background for the video frame, and then the foreground image of the hand-held probe can be obtained; Compare with the actual position for the probe need to be detected, the validity for detection will be evaluated. Experiment results show that the average accuracy for the validity detection in the video surveillance is about 93.26%, the recall rate is about 81.11%, and the F1 score is about 86.76% with the detection speed of 9.66 fps/s, which achieves the validity detection of the probe in video surveillance.