摘要： In recent years, the demand for IoT cameras has increased due to the high demand for image data. However, the image sensor is unsuitable as an energy-constrained edge device for IoT due to its high-power consumption. Therefore, periodic on–off scheduling of IoT cameras is a promising approach since video recording using image sensors is energy-intensive. Due to the constrained computing performance of edge devices, IoT is still based on cloud computing with energy leaks by transmitting all the data of edge devices to cloud. In this paper, we proposed energy-efficient communication via edge computing based on deep learning, which reduces power consumption by transmitting only images of interest classified using edge computing. We also designed and implemented CamThings, which is an energy-efficient IoT camera with periodic on–off scheduling and the proposed energy-efficient communication. To analyze and evaluate the efficiency of the proposed communication scheme, we implemented a power consumption model for CamThings. In an environment with a low interest ratio, the proposed CamThings is superior to the baseline method with only periodic on–off scheduling in terms of power consumption and lifetime. When the scheduling period T is 5s and the interest ratio is 0.1, the proposed method consumed 41% less power than the baseline method. As a result, CamThings has a lifetime of more than one month.