A sensor based system design for pose estimation and application

The real-time modeling of physical movements in digital environments has emerged as a growing field in recent years. Advances in electronics have significantly facilitated the transformation of physical reality into digital form. Digitally capturing the motion of human joint regions plays a critical role in both medical applications and systems synchronized with the human body. The precise and rapid transfer of joint movements to a digital platform enhances the accuracy and speed of medical diagnosis and treatment processes. Beyond medical fields, such systems offer significant advantages in areas such as human-robot interaction. Systems synchronized with the human body not only make physically demanding tasks more manageable but also reduce potential risks to individuals. This thesis aims to develop a system for predicting human body poses. The proposed system encompasses both electronic and software components. The electronic system is designed to be deployed across ten regions of the human body. Key components include a microcontroller, IMU sensors, a Wi-Fi module, and other electronic elements. Using IMU sensors, angular orientation data is collected from the human body and transmitted to a database via the Wi-Fi module. On the software side, a custom-designed interface retrieves sensor data from the database and processes it using an artificial neural network (ANN) model for pose prediction. The predicted pose is then displayed both visually and numerically on the same interface. The electronic components of the system are designed to be fully modular. A LiPo battery is used as the power source. The system also features a data generation mechanism for training the AI model. IMU sensor data is stored in the microcontroller's memory and used as input for AI model training. A dataset comprising 300,000 sensor readings was collected for training, resulting in a pose prediction accuracy of 90% for 10 distinct poses.