System Development Of Moving Human Body Detection Fusing PIR And Video Information

With social developing and people’s living standards improving, people requiremore safety in their house. Pyroelectric infrared (PIR) detector and video surveillancesystem are the most commonly used products in the field of home security. PIR detectorhas stable performance, low cost, no illumination requirement, meanwhile it’sconvenient to be concealed. However, there are serious false alarms and missed alarmsphenomenon which limits the further popularity of PIR detector. Video surveillancesystem is intuitive to use, proved to be effective to forensics, while video surveillancesystem needs a watchman to gaze the monitor constantly. With its noisy images in lowillumination area, video surveillance system is hard to achieve full-time monitoring.The main reason for false alarm and missed alarm is the lack of effective analysisof PIR signal. After years of study, our laboratory has developed a human bodydetection algorithm for PIR signal based on Autoregressive (AR) model. The simulationresult shows this algorithm can effectively reduce the false and missed alarm rate. So afurther verification in real system is needed. As the video surveillance system is difficultto distinguish human and non-human like leaves, fans, animals, etc. We introduced scalefeature and Hu moment features to traditional template matching algorithm and gotgood detection performance. So we designed and developed a system of moving humanbody detection fusing PIR and video information. The system is equipped with a PIRchannel and video channel. Because PIR channel performs stable under lowillumination, we fuse detection results from two channels with weighted voting methodbased on fusion theory and achieved full-time monitoring.The main works are as follows:1）Analysis the composition of the PIR detector and the characteristics of PIRsignal. Based on previous study, we developed a real-time acquisition device of PIRsignal and port human body detection algorithm based on AR model. The result showsthat it is effective to solve the traditional PIR detector’s false alarm and missed alarmsproblem.2）On the QXD-DM6446development platform, we developed a program to obtainPAL standard video signal, extract moving object, separate object from background. Weextract scale and contour feature from moving object. With an improved templatematching method we distinguish the moving human body or not. All the human detection processes are run on the DSP core of Davinci dual-core chip, according to thexDM programming framework.3）On the basis of human body detection of PIR channel and video channelrespectively, we fused the two channels detection results using voting method andobtained full-time monitoring ability.4）In order to record human intrusion evidence, we develop a program to encodeJPEG Image and store it to U disk. In order to observe the real-time status of the scenevisually, we port BOA open source project to build a web server and modifyWebCamServer open source project to build an image server, so that the real-timehuman intrusion image will be sent to PC browser.