The YOLOX target recognition framework and NVIDIA's Jetson development board are used as the core of the real-time edge calculation solution for the number of people in the elevator of industrial building exterior wall construction.A single fisheye lens is used for video signal acquisition,and the calculation results are fed back to the system through the RS-485 bus.The experimental results show that compared with the YOLOv3 algorithm,it ensures the overall system target recognition abil-ity,accuracy,and real-time,effectively reduces the system volume and calculation amount,improves the overall cost perform-ance of the system.In the actual environment,the refresh rate is 33 FPS,the recognition rate is 94.52％,the system volume is reduced by 70％,and the system power consumption is reduced by 95％.It can meet the application requirements of real-time identification of construction elevator scenes.
关键词
YOLOX/人数统计系统/施工升降机/嵌入式系统
Key words
YOLOX/people counting system/construction elevator/embedded system
维普
万方数据