摘要
为提升地铁站照明控制系统的灵活性,改善照明质量以满足不同乘客的视觉需求,提出一种基于乘客行为识别的地铁站照明控制方法。首先采用深度可分离卷积以及DenseNet网络构建深度可分离密连接网络(DSDNet),并利用网络嵌套的方式,设计基于YOLOV3-Tiny网络和DSDNet网络的行为识别算法;其次通过分区统计的方式实地调研地铁站乘客行为,并建立乘客行为数据集(PADS);最后依据行为识别算法和调研结果提出分区动态照明控制方法,并通过PWM调光方法进行LED灯具分组控制。实验结果表明:该算法在PADS数据集上的识别准确率达到97.472%,在UCF-101公共数据集上的识别准确率达到93.1%。使用OpenMv(可编程机器视觉模块)进行实物功能验证,证实通过识别人体行为可以准确地控制LED灯具的亮度;并借助DIALux evo照明设计软件进行照度、能耗分析,证实该方法可以较精准地调控相关区域的照度,与传统的群控法相比节电率可达到6.6%。
To enhance the flexibility of the lighting control system for subway stations and improve lighting quality to meet the visual needs of different passengers, a subway station lighting control method was proposed based on passenger action recognition. First, the deep separable convolution and DenseNet network were used to construct the deep separable densely connected network(DSDNet), and the network nesting method was used to design the action recognition algorithm based on the YOLOV3-Tiny network and DSDNet network. Second, field surveys of passenger action in metro stations were conducted through district statistics, and a passenger action data set(PADS) was established. Finally, based on the action recognition algorithm and the research results, a partition dynamic lighting control method was proposed, and the PWM dimming method was used to group and control the LED lamps. The experimental results show the effectiveness of this algorithm: The recognition accuracy rate on the PADS is 97.472%, and the recognition accuracy rate on the UCF-101 is 93.1%. Using OpenMv(programmable machine vision module) to carry out physical function verification, which confirmed that it can accurately control the brightness of LED lamps;and with the lighting design software DIALux Evo illuminance, energy analysis, confirmed that the method can be more accurate illuminance related regulatory region, with the conventional rate saving group control method can be compared to 6.6 percent.
作者
段中兴
丁青辉
王剑
李伟哲
DUAN Zhongxing;DING Qinghui;WANG Jian;LI Weizhe(College of Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China)
出处
《铁道科学与工程学报》
CAS
CSCD
北大核心
2021年第12期3138-3145,共8页
Journal of Railway Science and Engineering
基金
国家自然科学基金资助项目(51678470)。