Detection Of Two-wheeled Vehicles In Traffic Video Based On Convolutional Neural Network

In recent years,the intelligent transportation system has played a powerful regulatory role in China’s traffic management,which has aroused widespread concern in China.Vehicle detection technology is a research hotspot of intelligent transportation,and has achieved certain research results.The traditional vehicle detection method is to extract features by artificial rules,and is not suitable for complex and variable traffic environment.The vehicle detection method based on deep learning can automatically learn target features in various environments,which has stronger generalization ability and strong applicability to complex traffic environment.In the traffic system,the two-wheeled vehicles have more violations and are more likely to cause traffic accidents,so the detection of two-wheeled vehicles is imminent.Based on the research status at home and abroad,this paper will use the improved Faster R-CNN framework to detect twowheeled vehicles in traffic video.The specific research contents are as follows:1.Research on the detection technology of general vehicles and two-wheeled vehicles at home and abroad,and compare the traditional methods with the deep learning methods to fully understand their advantages and disadvantages.2.For the missed detection problem of the small-scale two-wheeled vehicle detection by Faster R-CNN algorithm,this paper makes the following improvements to the Faster R-CNN model: firstly,the size of the anchor box is modified for the two-wheeler data set,and secondly To compensate for the loss of the detail features of small target,the features of different convolution layers are merged when extracting features,thereby improving the sensitivity of the model to small-scale two-wheelers and improving the expression ability of the model.3.Aiming at the false detection of the small-scale two-wheeled vehicle detection by Faster RCNN algorithm,this paper introduces the hard negative sample mining strategy,and improves the recognition ability of the model through multiple training of hard samples,thus reducing false detection rate.4.The improved Faster R-CNN algorithm is verified by the two-wheel dataset of various real traffic scenes in this paper,and a variety of different detection methods are compared and analyzed to verify the effectiveness of the detection method of the two-wheel of traffic video.The experimental results show that the proposed method outperforms other methods in terms of detection accuracy and recall rate.