In this paper, we propose genetic algorithm-based pose estimation deep learning model optimization and anomaly detection for safety management in unmanned stores using radio frequencies. The radio frequency used is a millimeter wave in the 30 GHz to 300 GHz band. Due to the short wavelength and strong straightness, it is a frequency with less grayness and less interference due to radio absorption on the object. Using millimeter wave radar, we solve the privacy infringement problem that may arise from conventional CCTV image-based pose estimation and abnormal detection.
Deep learning-based pose estimation models generally use convolutional neural networks. The convolutional neural network is a combination of convolutional layers and pooling layers of different types, and there are many cases of convolutional filter size, number, and convolutional operations, and more cases of combining components. Therefore, it is difficult to find the structure and components of the optimal pose estimation model for input data. Compared with conventional millimeter wave-based pose estimation studies, it is possible to explore the structure and components of the optimal pose estimation model for input data using genetic algorithms, and the performance of optimizing the proposed pose estimation model is excellent. abnormal detection uses the key point of pose estimation. Keypoint-based activity recognition converts keypoint coordinates into sequence data and performs activity recognition through a cyclic neural network, using a type of cyclic neural network, a bidirectional short- and long-term memory neural network, and a circuit-type circulation unit to perform abnormal activity recognition in an unmanned store.
Data are collected for actual unmanned stores, and point cloud data and three-dimensional keypoint information of Kinect Azure are collected using millimeter wave radar for collapse and property damage occurring in unmanned stores. In addition, pose estimation and behavioral recognition were performed using 10 rehabilitation exercise data used in the MARS: mmwave-based Assistant rehabilitation System for Smart Healthcare study for accurate performance analysis.
As a result of the simulation, it was confirmed that the error was moored compared to the conventional pose estimation model, and the activity recognition result was also improved.