Multi-Channel Fusion Feature Vehicle Detection Based On Static Image

With the increasing number of vehicles,automobile safety has attracted more and more attention.How to improve the detection accuracy of the target vehicle on the premise of guaranteeing the detection speed is the current research focus.It can be known from many research literatures that it is impossible to solve the scene with high complexity by using a single feature.To solve this problem,this paper uses a multi-channel fusion feature method to detect the target vehicle.The main research contents of this paper are as follows:1)This paper mainly studies the development status of vehicle detection and identification technology at home and abroad,and analyzes the advantages and disadvantages of vehicle vision processing system,analyze and compare their advantages and disadvantages based on prior knowledge,machine learning and deep learning.Finally,the vehicle detection algorithm is improved based on machine learning.2)Vehicle extraction features in vehicle detection are discussed in depth,and a variety of vehicle extraction features are analyzed and compared.Finally,multi-feature extraction and serial fusion are adopted to extract the feature vectors of the target vehicle,thus,multichannel fusion features are constructed.In order to reduce the dimension of fusion features,the maximum pooling method is adopted,the computational amount of the algorithm is greatly reduced,and the training speed and detection efficiency are greatly improved;3)In the process of vehicle target feature extraction and verification,multi-channel fusion features are applied to the region of interest of the vehicle,the result shows that the multi-channel fusion feature has lower error rate and higher detection rate than the single feature.Through the combination of color LUV channel,it also has better robustness to complex light changes.The Adaboost classification was improved,and Adaboost detector was combined with HOG and Haar features to form an algorithm of multi-feature cascade detector to make up for the defect of missing single feature detection.