UAV Visual inspection System Based On Deep Learning

• Mo Yang

In public events such as music festivals or large-scale gatherings, expect the event itself, the most intuitive feelings are crowded people and nowhere parked private cars. As organizer of the activities and police, how to solve the lack of security personnel, how to effectively conduct patrols and calculate the number of people at each location to divert the crowded areas and prevent stamping accidents, how to grasp the information encountering unexpected situations or accident scenes and designing scientific and effective programs in the shortest time. The core of those problems is to get information and analyze it quickly and efficiently. As a solution, we plan to use drones with certain artificial intelligence functions as auxiliary means. This project is about an unmanned aerial vehicle(UAV) target detection system based on deep convolutional neural network(DCNN). The project is divided into two parts: the front end is the UAV information collection terminal and the back end is the information analysis system and flight control terminal. The UAV transmits data to the back end through 4G network or Wi-Fi. We set up a virtual LAN by creating a VPN server on the onboard Raspberry Pi to realize data and image transmission communication at the front and back ends. The back end analyzes the information in real time through a well-trained deep learning model based on DCNN to obtain the result. This project adopts DCNN feature extraction and recognition technology and incorporates OpenCV face algorithm to detect and count targets crowds and vehicles. Mean square error and mean absolute error are used as the evaluation criteria for model accuracy to measure the difference between the predicted value and the real value. In this project, we will launch drone to automatically take off, land and identify people and vehicles with an accuracy of more than 85%.