基于深度图的驾驶舱内飞行员动作识别

Pilot Action Identification in the Cockpit

驾驶舱内飞行员工作量的计算需要飞行员的动作数量、动作时间等动作信息。目前,关于动作识别的研究一般都是对特定动作,如走、跳等的识别,无法应用于驾驶舱内的飞行员。同时,由于飞行员操作基本由手来完成,因此对飞行员的动作识别基本可以认为是对手部动作的识别。据此提出一种基于深度图的飞行员动作识别方法,该方法先通过对飞行员手部进行跟踪,再通过基于动作段的方法确定飞行员动作。此外还提出一种触发方法,以实现系统对动作的自动识别。实验结果显示,所提方法的动作识别率约为94.06%,表明该方法能够有效地识别飞行员动作。

The identification of the pilot actions in the cockpit is of significant importance for the accurate calculation of the pilot＇s workload. At present, the research about action identification generally focused on special actions, such as walking and jumping, which is not applicable for the pilot inside cockpit. As the pilot actions are basically implemented by the hand, the pilot action identification can be considered as the pilot hand movement identification. A novel technique is proposed here for the identification of the pilot actions based on the depth information, which tracks the pilot＇s hands at first, and then identifies the pilot action. In addition, an approach to activate the algorithm is also presented for automatic recognition of the actions. The results of the experiments show that the identification rate of the technique is about 94.06%, indicating that it can precisely identify the pilot actions in the cockpit.