Construction Vehicle Detection And Tracking Based On Deep Learning

Construction vehicles are important means of production in engineering construction,and the rise of monitoring systems has brought many conveniences to the monitoring and management of construction vehicles.Traditional algorithms have problems such as insufficient accuracy and poor real-time performance when detecting and tracking engineering vehicles.This topic focuses on the application of deep learning algorithms in the detection and tracking of construction vehicles.Finally,the design and development of the construction vehicle detection and tracking system is completed according to actual needs.Intelligent management of construction vehicles can save labor costs and improve the safety of the construction process.It also contributes to the digital construction of project management and improves work efficiency.The main contents of this thesis are as follows:The traditional target detection algorithm and the engineering vehicle detection algorithm based on deep neural network are studied.The image is preprocessed using traditional image preprocessing algorithms such as image denoising,enhancement,and image edge detection,and then the experiments are carried out separately.The results show that the use of HOG+ SVM’s traditional target detection algorithm has low detection accuracy,while Faster-RCNN detection algorithm has poor real-time performance.Afterwards,the YOLOv5 algorithm with high detection accuracy and real-time performance is used to detect the construction vehicles.Aiming at the phenomenon of the prediction frame position deviation and target missed detection in the detection,the loss function and the NMS method are improved experiments.The improved algorithm improves the average detection accuracy,reduces the missed detection phenomenon,and accelerates the convergence speed of the loss function.An engineering vehicle tracking algorithm is designed that uses an improved detection algorithm as a detector,uses Kalman filter to estimate and updates,and uses Hungarian algorithm to correlate data between consecutive frames.In view of the problem that the target identity information is switched frequently when the camera is moving and the vehicle is at a non-uniform speed,the deformed GIo U is used to calculate the Intersection over Union between the trajectory and the detected target prediction frame,which reduces the IDS in the tracking process.Finally,a construction vehicle inspection and tracking system is designed,and the algorithm is applied to production practice.