室外人体脚步声事件及环境联合识别

为了实现室外人体脚步声事件及环境联合识别,首先,设计一个复杂相似环境下的人体跑动和行走数据集,提出一种交叉双脚步声的分割方案,对连续脚步声信号进行交叉分割;然后,从事件和环境的角度分别提取特征,并从任务平衡的角度设计两种融合特征;最后,采用3种深度学习模型对任务进行精确地识别.结果表明:文中方法简化平衡了任务,使室外人体脚步声事件及环境联合识别的多任务设计不需要复杂模型就能实现精确识别.

Outdoor Human Footsteps Event and Environment Joint Recognition

In order to realize the joint recognition of outdoor human footsteps events and environment, firstly, a human running and walking data set in a complex and similar environment was designed, and a cross double footsteps segmentation scheme was proposed to cross segment the continuous footsteps signals. Then, features were extracted from the perspectives of events and environment, and two fusion features were designed from the perspective of task balance. Finally, three deep learning models were used to identify the task accurately. The results showed that the proposed method simplified and balanced the tasks, and the multitask design of joint identification of outdoor human footsteps events and environment could realize accurate identification without complicated models.