Automatic Measurement of Human Ear Parameters Based on Object Detection and Landmark Extraction

How to efficiently obtain human ear parameters is one of the key challenges in personalized HRTF (Head-Related Transfer Function). Traditional methods using 3D laser equipment for human ear parameter measurements require significant manpower and cost investment, while edge detection and contour extraction suffer from a large margin of error in measurement accuracy. To address these challenges, we propose an automatic measurement method for human ear parameters based on object detection and landmark extraction only from a side face image and a frontal ear image. Firstly, to address the issue of scale variations in human ear parameter measurements, a side face image containing both the iris and the human ear is collected. MediaPipe FaceMesh is used for iris detection and an improved YOLOv5s model is used for human ear detection. By utilizing the approximate invariance of human iris diameter, the actual height of the human ear is obtained. Secondly, to calculate human ear parameters more accurately, a frontal ear image is collected, and 55 landmarks of the human ear are extracted. Suitable landmarks are selected to calculate the pixel distances of ear parameters. Finally, based on the actual height of the human ear and the pixel distances of ear parameters, the actual distances of ear parameters are calculated. Experiments show that the margin of error of the proposed automatic measurement method for human ear parameters is about 6%, which meets the requirements of practical applications. Our code and pre-trained models are available at https://github.com/zzb010221/Automatic-measurement-of-human-ear-parameters.