基于四摄像头的全景目标检测系统

Panoramic target detection system based on quad camera

目前基于图像的检测系统多以单摄像头为主,但是单幅图像的视野有限,往往存在检测盲区,为此,设计了一种基于四摄像头的全景目标检测系统。首先搭建了由3B+树莓派、摄像头、LCD液晶屏、移动电源联合组成的系统硬件,基于QT设计了可进行检测结果显示与参数设置的用户交互软件界面;系统通过四部摄像头实现对周边360°场景进行取景,然后对采集到的图像进行预处理与拼接,最后通过基于EfficientDet深度学习模型检测器进行目标检测与结果显示。实验结果表明,该系统可便携移动,能实现360°全景取景与目标检测功能且目标检测准确率达94%以上。

At present,most of the image-based detection systems are mainly single-camera,but the field of view of a single image is limited,and there are often blind spots,so a panoramic object detection system based on four cameras is designed in this paper.Firstly,the system hardware composed of 3B+Raspberry Pi,camera,LCD screen and mobile power supply was built,and a user interactive software interface for detection result display and parameter setting was designed based on QT.The system uses four cameras to frame the surrounding 360-degree scene,then preprocess and stitch the collected images,and finally perform object detection and result display through the EfficientDet-based deep learning model detector.Experimental results show that the system can be portable and mobile,and can realize 360-degree panoramic framing and target detection functions,and the accuracy of target detection is more than 94%.