Abstract
This paper investigates the fixed-time sliding mode control (FTSMC) problem for a quadcopter unmanned aerial vehicle (QUAV), which is driven by a lip-reading recognition module. The lip-reading recognition module is consisted of a trained deep neural network with the structure of 2D-Conv + GhostNet + TCN. In order to reduce the communication burden between the remote controller and the QUAV as well as reduce the computation burden in running the lip-reading recognition module, the event-triggered mechanism is introduced to the position controller design. The low-bound of the triggering interval is derived explicitly so that the Zeno phenomenon can be excluded. Furthermore, in order to overcome the main obstacle in high-accuracy control of QUAV, this paper launches a novel wind disturbance rejection approach by using wind field model, which is motivated by the physical dynamic characteristics of the practical wind. Specifically, the wind disturbance is estimated in the designed FTSMC by applying a specific wind field equation with preassigned physical parameters. To further reduce the chattering in the controller, a fitting technique is introduced via a local multivariate linear regression. Finally, both simulation and human-in-the-loop experiment results verify the applicability of the proposed control approach for the lip-reading-driven QUAV system. Note to Practitioners -This research is motivated by the need to design lip-reading-driven QUAV. In noisy environments or when silence is required, the efficiency of traditional human-computer interaction methods such as speech recognition is greatly reduced. Especially for people with damaged vocal cords, speech recognition is not achievable. In addition, it is difficulty to realize high-precision anti-interference control of QUAV with lower computational and communication burdens. In order to solve these problems, this research designs a lip-reading recognition module for QUAV control to cope with various complex application scenarios and realizes high-performance control by FTSMC algorithm. The key of this work to save system resources is to introduce the event-triggered mechanism into the position controller of the QUAV. In addition, this paper introduces the wind field model into the QUAV model to realize the wind disturbance suppression. The lip-reading-driven QUAV proposed in this paper have a wide range of applications, such as controlling QUAV in hazardous environments and improving the efficiency of interaction between human and QUAV.
