Enhancing Image-based Positioning With a Novel Foot Position Extraction Algorithm and Machine Learning

In recent years, the rapid advancement of deep learning has enabled researchers to locate individuals indoors using image recognition techniques. As a result, we propose an indoor positioning method that combines image and machine learning techniques. This study utilizes the robust OpenPose pose estimation model to accurately identify human joints and extract the position of the individual's foot from an image. Then, we apply the 2D direct linear transformation approach to determine the coordinates of the individual's foot position in the indoor environment. To reliably determine the position of the foot in complex indoor spaces where potential obstructions often occur, we propose a novel foot position extraction algorithm. Finally, we employ classical machine learning models, such as linear regression, to reduce distance errors caused by lens distortion and other factors. Our proposed method achieves highly accurate indoor positioning with an average distance error of only approximately 0.4 meters, just by using a single camera.