Design Of The Modular Reconfigurable Flight Array Target Visual Tracking System

Unmanned aerial vehicles(UAVs)equipped with target tracking systems are extensively utilized in critical areas like emergency security,traffic law enforcement,and pedestrian tracking.However,the majority of current UAVs have fixed configurations,limited environmental adaptability,and struggle with obtaining target coordinates and maintaining stability during target tracking tasks.These issues present significant challenges for UAVs during target tracking tasks in various scenarios.To address these issues,this paper focuses on the development of a modular reconfigurable flight array target tracking system,which can adapt to different flight tasks or scenarios by utilizing an improved target tracking algorithm.This study aims to investigate the system’s capabilities from various aspects to enhance its effectiveness in target tracking:(1)Design of a modular reconfigurable flight array platform.To address the limitations of current fixed and single-functional UAVs,a modular design approach was introduced,which includes a single-rotor flight module,a visual module,and a mechanical connection structure for assembling various modules.The overall structure of the flight array was optimized through prototype production and experimental analysis.Additionally,the results of the structural design were utilized to select the power system for the flight array,ultimately leading to the successful design of a modular reconfigurable flight array platform.(2)Configuration recognition and system modeling of the flight array through visual recognition.To address the challenges of autonomous ground assembly and controller matching of the flight array,a target detection algorithm was utilized to recognize and solve the coordinates of the flight array from an aerial perspective.This provided critical information on the coordinates of each flight module.Using the Newton-Euler equation and translation axis theorem,a generalized flight array dynamic model was constructed,and the flight module coordinates were incorporated into the model for online control allocation.Finally,through simulation experiments,the effectiveness of the proposed method was verified.(3)Design of a multi-target tracking algorithm.To address the limitations of the Deep SORT algorithm in small onboard computers,such as slow detector operation,delayed transient response,and the inability to specify a single target,an improved Deep SORT multi-target tracking algorithm was proposed.First,a Ghost lightweight network module and NAM attention mechanism were introduced into the detection algorithm,which effectively reduced the parameter count and improved the detector’s transient performance without compromising its detection accuracy.Then,a single target selection strategy is introduced to enable the flight array to continuously track a single target of interest.Additionally,PID control and sliding average filtering were used to smooth the control commands obtained by the target tracking algorithm,thus mitigating the shaking problem during the flight array’s target tracking process.(4)Experimental validation of the modular reconfigurable flight array target tracking system.Flight experiments were conducted indoors and outdoors to test the flight array’s stability.Outdoor experiments were performed to track pedestrian targets using the flight array’s improved algorithm.The comprehensive performance of the improved algorithm was evaluated during the tracking task on the flight array.The results demonstrated the feasibility of the designed system in this paper.