关键词: Jansen机构, 运动曲线, 非结构环境, 仿生度, 深度学习, 特征融合, 目标检测, 损失函数

Abstract:

In view of problems such as high training cost for guide dogs, this study proposes a walking mechanism with visual recognition function, which can realize a single stepper motor to drive a four-legged mechanism to walk. Through the simulation of the movement of the walking mechanism, the obtained data show that the four feet can alternately rise and fall relative to the ground, and can run smoothly in a more complex non-structural environment. Taking the recognition of the traffic lights on the crosswalk as an example, a visual recognition algorithm for mechanical guide dogs based on YOLO is proposed to determine whether the current signal status is passable. The mechanical guide dog recognizes 87.96% of the green light and 89.38% of the red light, and the detection speed is 23.5 frame·s^-1,indicating that the mechanical guide dog can detect and distinguish the traffic light information of pedestrian crossings in real time.

Key words: Jansen mechanism, motion curve, unstructured environment, bionic degree, deep learning, feature fusion, target detection, loss function