| With the swift and violent development of computer vision technologies,machine vision detection technologies based on neural network have made a breakthrough,in which traffic target detection algorithms based on deep learning have gradually entered the vision of the majority of researchers.However,the changing time,changeable weather environment and various types of targets pose severe challenges to multi-target detection(especially accurate detection)in intelligent transportation scenarios.Therefore,based on an method of three-dimensional scenario space,this thesis uses the deep convolution neural network model YOLOv3 as the basic model,and proposes two accurate detection algorithms to study machine vision detection technologies in multi-objective scenarios of intelligent transportation.The main works are as follows:I.Firstly,intelligent transportation systems and machine vision technologies are summarized around traffic target detection algorithms.Then,deep convolutional neural networks,classical object detection models and object detection evaluation indexes are introduced in detail.II.A priori-driven traffic complex multi-objective adaptive and accurate detection algorithm is proposed: Firstly,target detection models for three-dimensional scenario are constructed to realize the accurate detection of a certain kind of target in a specific time and weather scenario.Secondly,the multi-dimensional model is selected according to the prior knowledge,so as to accurately detect a certain type of target in a specific time and weather scenario.Finally,the multi-target adaptive joint detection(i.e.multi-dimensional models fusion)is carried out.The multi-target adaptive joint detection mainly includes the process of multi-dimensional models fusion and the threshold adaptive adjustment mechanism,so as to realize the accurate detection of multi-target in complex scenarios of intelligent transportaion.Through experiments,the priori-driven traffic complex multi-objective adaptive and accurate detection algorithm is compared with YOLOv3 algorithm.The experimental results show that the proposed algorithm can realize the accurate detection of multi-objective in complex scenarios of intelligent transportaion,and the evaluation indexes are improved.III.A stratified and hierarchical joint accurate detection algorithm for traffic complex multiple targets is proposed: Firstly,three types of target detection models for single-dimensional scenario are constructed,which reduces the cost compared with the construction of target detection models for three-dimensional scenario.Secondly,the multi-objective stratified and hierarchical joint detection is carried out.The multi-objective stratified and hierarchical joint detection mainly includes the stratified joint detection process and the hierarchical joint detection mechanism,so as to realize the accurate detection of multi-target in complex scenarios of intelligent transportaion.Through experiments,the stratified and hierarchical joint accurate detection algorithm for traffic complex multiple targets is compared with YOLOv3 algorithm.The experimental results show that the proposed algorithm can realize the accurate detection of multi-objective in complex scenarios of intelligent transportaion,and the evaluation indexes are improved. |
