Real-Time Ear Detection Based On Embedded Systems
Yuan Li, Fei Lu · 2018
Ear detection is the first step of ear recognition. This paper proposes a real-time ear detection system based on embedded systems. An improved YOLO network is proposed for ear detection. With the same network depth, the width of the improved yolov2-tiny network in YOLO has been reduced to a quarter of the conventional network. By introducing batch processing and reducing the regularization coefficient of the improved yolov2-tiny network, the detection accuracy has been improved and the detection time has been reduced. The proposed model is more applicable for embedded implementation. On Nvidia Jetson tx2 real-time detection of the ear has been realized. The image is captured by an external camera and the ear will be marked when the ear appears in the image. On Nvidia Jetson tx2, the real-time detection frame rate is 30 frames per second which can meet the real-application requirements.