基于SSD网络的电梯内电动自行车检测研究

Research on detection of electric bicycles in elevator based on SSD network

为减少因电动自行车违规操作而造成的消防安全事故,杜绝电动自行车进电梯的违规行为,基于深度学习SSD目标检测网络,使用VGG16、EfficientNet、MobileNet 3种主干网络,研究SSD网络对电梯内电动自行车检测的可行性,分析比较3种网络的检测效果,并提出基于双摄的检测方法,进一步提高电梯场景下检测准确度,减少误检误报警。研究结果表明:SSD检测网络对电梯内电动自行车检测效果良好,其中SSD_MobileNet网络更适用于工业领域,双摄检测方法的检测准确率均大于90%。

In order to reduce the fire safety accidents caused by the illegal operation of electric bicycles and prevent theelectric bicycles from entering elevator,based on the deep learning single shot multibox detector(SSD)target detection network,three backbone networks of VGG16,EfficientNet and MobileNet were used to study the feasibility of applying SSD network on thedetection of electric bicycles in elevator.The detection effect of the three networks were analyzed and compared,and a dual-camera detection method was proposed to further improve the accuracy of detection in the elevator scene and reduce the false detection and alarm.The results showed that the SSD detection network had good effect on the detection of electric bicycles in the elevator,among which the SSDMobileNet network was more suitable for industrial application,and all the accuracy of dual-camera detection method reached 90%.