CMSS: Use Low-power IoT Cameras to Monitor Store Shelves
Xiaotian Fan, Yubo Yan, Panlong Yang, Feiyu Han · 2021
In this paper, we propose a method of using low-power small cameras to detect shelf status. For shopping malls, especially large shopping malls, when to replenish the shelves is very important. Whether they can replenish them immediately or not is directly related to their turnover. Now most shopping malls use manual inspection mode, that is, a waiter is responsible for certain shelves. They will replenish shelves in time when they are need. But this method wastes too much manpower. In our scheme, we use low-power IoT C ameras to M onitor S tore S helves (CMSS). Firstly, we put a specific photo on the back of the shelf and put the camera in front of the shelf to face the photo. When the goods on the shelf are emptied, the camera recognizes the graphics of the photos posted on the shelf in advance, and will send a message to inform the manager that replenishment is needed. For image recognition, we use an ultra low-power FPGA chip and run a convolutional neural networks(CNN) model on it to achieve. The chips and cameras we use are low power consumption. Their power consumption is milliwatts level, which is also the lowest power consumption device we know of running CNN. The power consumption of the entire system is less than 6 mW, and the price is about 22 dollars. Our experimental results show that on a shelf with a width of 40cm, the maximum lateral distance that the system can recognize is about 40cm. Compared with the existing solutions, our solution has the advantages of lower cost, low power consumption and easy maintenance.