用于城市流浪猫生态考察的远程视觉监测系统

Remote Vision Surveillance System for Ecology Investigation for Urban Stray Cats

针对城市流浪猫的考察与救助等问题,开发了一种基于机器视觉和远程通信的监测系统。其移动端和上位机分别部署于现场和监控中心。移动端通过摄像头采集流浪猫图像,采用多尺度块局部二值模式(MB- LBP)检测猫脸纹理,运用Haar- Like算子描述猫脸眼部特征。综合以上2种特征,构建并运用Gentle AdaBoost算法训练2级级联的分类器;移动端与上位机间的图像传输可根据现场蜂窝网覆盖情况灵活选择3G或LoRa通信模式。基于1 000张图片的猫脸检测结果,正确率(TPR)为70%,误检率(FPR)为28%。样机的现场测试结果验证了系统各项功能的有效性,将机器视觉应用于城市生态管理具有积极意义。

A monitoring system based on machine vision and remote communication has been developed for the investigation and rescue of urban stray cats.The mobile terminal and the host computer are deployed in the field and monitoring center respectively.The mobile terminal detects images of stray cat by utilizing camera.The fur features of cat face are detected by multi-block local binary pattern (MB-LBP),and the ocular region characteristics of cats are described by the Haar-Like operator.Combining two characteristics above.the Gentle AdaBoost algorithm is built and used to train a 2-stage cascaded classifier.The suspected images are uploaded to the host computer via telecommunication,either 3G or LoRa,based on cellular network conditions on the spot.Computational results on 1 000 images of cat face from the benchmark dataset demonstrate the true positive ratio (TPR) and false positive ratio (FPR) of cat-face detection are 70% and 28%,respectively.Field tests on prototype verify the overall effectiveness of the whole system.This surveillance system will play an active role in applying the technology of machine vision to urban ecology management.