Research On Engineering Vehicle Detection And Vehicle Type Recognition Under Construction Scene

In the last few years,with the rapid development of national urbanization,a large amount of land has been developed,and construction phenomena is common,large-scale engineering vehicles such as cranes and diggers are everywhere.The vehicles of this special type of engineering vehicles are very destructive.If you don't pay attention,it will cause irreparable damage to people or things around you,causing huge economic losses.The use of intelligent video image monitoring technology to detect and identify the condition of the vehicle can prevent the occurrence of danger in a timely and efficient manner.The key technology is vehicle detection.The vehicle detection technology is applied to detect the engineering vehicle to realize the monitoring demand for the engineering vehicles parked or operated in the construction scene,accurately warning,reduce the cost and prevent the danger,and provide protection for the personal safety of the workers.This paper focuses on the detection and recognition technology of engineering vehicle based on deep learning.It mainly solves two kinds of key problems: one is the problem of scale change caused by the difference of shooting angle and height in the complex background,and the other is engineering vehicle's parts are flexible,causing a problem of large deformation of the appearance.Therefore,the main work of this paper is as follows:(1)The engineering vehicle images of different types and different shapes taken at different angles will be collected from various construction scenes to form a sample set,and the engineering vehicle data set will be created with reference to the standard data set format.Moreover,in order to prevent over-fitting in the training model under the condition of a small number of effective samples,the data enhancement technology is used to expand,increase the diversity of the data set,and improve the generalization ability of the model.(2)Aiming at the problem that the candidate frame with variable size and fixed size in the image can not fully adapt to the target,an adaptive vehicle-based detection and vehicle identification algorithm based on adaptive search is proposed.By using the adjacency and scaling network to adaptively generate candidate suggestion regions,it is better to process targets with different scales in an image,and add an online difficult case mining algorithm in the detector to solve the problem of uneven sample categories and improve the model.The generalization of the training increases the effectiveness of the training.The average accuracy of the engineering vehicle dataset is 91.6%.(3)An engineering vehicle detection and vehicle identification algorithm based on global and local convolution feature fusion is proposed,which can solve the problems of different scales and complex deformations in the detection and recognition of engineering vehicles.By establishing a multi-scale region of interest pooling layer to obtain the overall structure and context information of the image,the global features of the target are extracted and utilized.The localized feature of the location-sensitive region of interest extracts the local features of the target,and the global and local features are weighted and fused at the fully connected layer.Finally,the vehicle position and category are jointly predicted by multi-task learning,and the average accuracy average is 92.5%.The global and local features have better complementarity.