基于图像识别与信息融合的电梯智能阻车系统研究

Research on intelligent Vehicle Blocking System of Elevator Based on Image Recognition and Information Fusion

使用电梯智能阻车系统可有效防范火灾隐患。从图像识别和信息融合两个角度入手,开发电梯智能阻车系统。监控视频信息采集方面,提出外景抓拍检测加内景摄像头二次识别确认的方式,可减少遮挡或伪装电动车进入电梯造成的误判和漏判。在图像识别上,通过试验法现场拍摄电动车进入轿厢的各种位姿,辅助以网络爬取扩充样本数量,并依据局部不变性原理,进行缩放、旋转、镜像,训练构建模型库;使用嵌入式处理器采用YOLO V3深度学习的目标检测算法,完成对内外摄像头抓拍输入图像的识别、推理,自动识别电动车。信息融合方面,加装传感检测单元以提高目标识别的准确率,在电梯轿厢底部合适位置布局地感线圈检测电动车金属底架。最后,控制模块根据融合识别结果给出控制策略。发现电动车进入电梯时,执行模块控制电梯开门按键常通,无法关门,同时发出声光报警。结果表明,与单一轿厢内部摄像头检测相比,所提出的方法识别准确率得到有效提高。

The use of elevator intelligent car blocking system can effectively prevent fire hazards.Based on image recognition and information fusion,an elevator intelligent vehicle blocking system is developed.In the aspect of surveillance video information acquisition,the method of exterior shot detection and interior camera secondary recognition and confirmation is proposed,which can reduce the misjudgment or omission caused by blocking or camouflaging electric vehicle entering the elevator.In terms of image recognition,various positions and poses of the electric vehicle entering the car are photographed on the spot through the test method,and the number of expanded samples are obtained with the aid of network crawling.Moreover,according to the principle of local invariance,the model library is trained and constructed.The embedded processor adopts YOLO v3 deep learning target detection algorithm to complete the recognition and reasoning of the input images captured by internal and external cameras and automatically identify the electric vehicle.In terms of information fusion,a sensing and detection unit is added to improve the accuracy of target recognition,and the ground sensing coil is placed at the right position at the bottom of the elevator car to detect the metal chassis of the electric vehicle.Finally,the control module gives the control strategy according to the fusion recognition results.It is found that when the electric vehicle enters the elevator,the execution module controls the elevator door opening button and can not close the door.At the same time,the sound and light alarm is issued.The results show that compared with the single elevator capsules internal camera detection,the method effectively improve identification accuracy.