Design And Implementation Of Camera Drone And Face Detection System Based On Domestic Master Chips

With the development of microelectronic technology and control theory,micro quad-rotors camera drones have the characteristics of low power consumption and high mobility.They can achieve controllable flight and aerial photography,and play an important role in fields such as aerial reconnaissance and dynamic security.With the development of artificial intelligence,lightweight face detection algorithms based on convolutional neural networks demonstrate performance advantages.They can predict face regions in images quickly and accurately,and produce a marked effect in fields such as identity recognition and security monitoring.By applying the face detection algorithm to aerial images of the drone,we can realize face detection from the perspective of the drone.Most of the available drones adopt foreign chip solutions,which leads to the high cost due to the chip shortage.Compared with domestic chips,foreign chips have no price advantage.In addition,many state-of-the-art lightweight face detection algorithms have low runtime efficiency on low-performance devices such as portable computers,and their actual performance needs to be improved.For these reasons,this thesis constructs a camera drone and face detection system based on domestic master chips,which involves the design of micro quad-rotors camera drone,the research on lightweight face detection algorithm and the design of ground control software.The system realizes the effective application of face detection in aerial photography scene.The main works of this thesis include:(1)We design a micro quad-rotors camera drone based on domestic master chips,which is composed of flight control software and hardware,wireless communication and image transmission software and hardware,structural components and parts.Based on CW32F030C8T6,the flight control software and hardware are designed.With MPU6050 and BMP280 sensors,we can realize attitude and height control by developing drivers and transplanting open-source flight control algorithms.Based on ESP32-WROVER-E,the wireless communication and image transmission software and hardware are designed.With OV2640 camera,we can achieve command information receiving-sending and image transmission by applying the dual-core multi-task architecture and open-source class library.Through designing the rack and battery compartment,selecting motors,blades and battery reasonably,we can accomplish the structural fixation and overall assembly of the drone.The hardware debugging and functional testing results show that our designed drone is eligible for flight and aerial photography,and has a high cost performance.(2)We study a lightweight face detection algorithm based on dynamic equilibrium matching strategy.Our proposed face detection algorithm applies lightweight convolutional neural network structure,takes MobileNet series as the backbone,and uses the dilated encoder to enhance the single-level feature.By developing dynamic equilibrium matching strategy,we can effectively solve the imbalance problem on anchors in the training phase.The experimental data show that compared with existing algorithms,our studied face detection algorithm achieves faster detection speed,lower computation and model capacity while assuring high detection accuracy.(3)We develop a ground control software based on Qt to control and monitor the designed drone,and apply the studied face detection algorithm to carry out real-time face detection on aerial images.Combined with design of visual interfaces and database,the software realizes drone control and monitoring,aerial video stream processing,lightweight face detection and other functions,and can run smoothly on low-performance devices.Taking a low-performance laptop as the ground terminal,we conduct system test and verification.The results show that our constructed camera drone and face detection system can realize drone flight control and monitoring,real-time aerial photography and face detection with low cost and high reliability through the efficient cooperation between the micro quad-rotors camera drone and the ground control software deployed with the lightweight face detection algorithm.