Highway Parking Event Detection Based On Improved Haar-like+adaboost

Parking event detection is an important part of the highway monitoring system.Compared with other parking event detection algorithms,Haar-like+Adaboost-based target detection algorithm has the following advantages: the target classification ability of the algorithm is better,the demand for image samples is less,and the feature training process is not easy to over-fitting.This method has achieved good results in the detection of parking events.However,the detection of parking events in complex scenes of highway sections faces disturbances caused by illumination,viewing angle and scale changes,which makes the detection algorithm based on Haar-like+Adaboost unable to guarantee the real-time performance and detection accuracy of the algorithm.Therefore,it is of great theoretical and practical significance to deeply analyze the characteristics of parking scenes in highway sections,analyze the shortcomings of the application of algorithms in highway scenarios and improve them to improve the performance of detection algorithms to meet application requirements.Firstly,the paper analyzes the problems and difficulties of vehicle image feature description method and detection algorithm based on Haar-like+Adaboost applied to highway sections.The Haar-like feature extraction mechanism of vehicles under complex conditions and the improved Adaboost cascade parking event detection model are studied.Finally,a set of highway parking event detection scheme based on Haar-like+Adaboost is formed.The main research contents of this paper include:(1)Research on Haar-like feature extraction method for vehicles under complex conditions.Aiming at the Haar-like feature extraction problem under the change of viewing angle,the proposed method is to flip the potential target area of the image through the automatic recognition of the region of interest and the mapping of the two-dimensional image,which reduces the significant change of the appearance characteristics of the vehicle.Aiming at the Haar-like feature extraction problem under illumination variation,the proposed method is to perform real-time luminance detection and adaptive Gamma transformation on the background image obtained by single Gaussian background modeling,which reduces the vehicle target luminance variation caused by frequent illumination changes.Aiming at the Haar-like feature extraction problem under scale change,the proposed method is to improve the original fixed-scale feature mapping method to the adaptive scale feature mapping method.This method reduces the Haar-like feature mapping bias caused by the scale change.The experimental results show that the above improvement effectively reduces the intra-class differences of Haar-like features of vehicles in multi-variable scenarios,which is beneficial to improve the recognition accuracy and generalization ability of detection algorithms.(2)Improvement of the cascading classifier detection model under the Adaboost framework.Aiming at the problem that the background difference method has insufficient ability to represent the contour features of the vehicle and the local spot interference caused by the lack of foreground target integrity,such as shade,cloud shadow and water stain,this paper presents a candidate region extraction scheme.By comparing the difference between the foreground and the background image of the canny texture information,the accuracy of the potential target candidate region is improved.On this basis,in view of the time-consuming problem of Adaboost algorithm,the proposed method is to eliminate the atypical Haar-like features that do not have better classification ability,and improve the training speed under the premise of ensuring the training effect.Finally,according to the characteristics of the local distribution of vehicle features in the highway scene,this paper adds a variable-scale Haar-like center feature based on the original algorithm.These features are involved in the Adaboost-based feature screening process to improve the classifier model’s ability to classify vehicle targets and non-vehicle targets.Finally,after the above two aspects of research,this paper uses the C++ language to implement the improved Haar-like+Adaboost-based parking event detection method on the Windows platform,and uses the historical video collected by the highway monitoring system to make the sample and Sample training.Finally,a rigorous comparison test was conducted.The experimental results show that the improved method can improve the representation ability of Haar-like features in highway complex environment,and improve the accuracy of parking event detection under the premise of ensuring the real-time performance of the algorithm.