Research On Pedestrian Detection Of ADAS System Based On Monocular Vision

With the development of science and technology,the application of artificial intelligence technology has become more and more extensive,and each industry has set off a wave of artificial intelligence.In advanced driving assistance system,how to reliably and accurately detect pedestrian targets in traffic environment has become the main focus of attention.In order to ensure the safety of everyone on the road to the greatest extent,the driver assistance system has more stringent standards for the detection speed and the accuracy that can be achieved by the pedestrian detection algorithm.Common detection methods can meet the requirements of detection speed,but they cannot meet the accuracy requirements.In order to solve the above-mentioned problems,this thesis adopts the pedestrian detection algorithm based on deep learning to study the pedestrian detection problem.The main research contents of this thesis include the following aspects:(1)The research status of pedestrian detection and recognition at home and abroad is analyzed.This thesis made a summary about the existing pedestrian detection,proposed the existing pedestrian detection problems existing in the research are: test results with the Angle of background,illumination,and changes in the environment changes,the change of the scale and obscured factors will affect the final results,the influence factors of test results received more.After that,the main research contents of the thesis are put forward.(2)Analyze the existing basic theories of pedestrian target detection.Analyzes the international commonly used on pedestrian detection data set in the existing data sets to add some pictures as a data set in this thesis,the current widely used for feature extraction and classification algorithm was analyzed,and points out the advantages and disadvantages of each,at the end of the maximum inhibition in the field of target detection and region recommendation algorithm is discussed and analyzed.(3)The advantages of the Faster RCNN pedestrian detection algorithm are analyzed.The feature extraction network and regional recommendation network of Faster RCNN are built.In order to reduce the error of the algorithm,the network loss function is established.On this basis,the back propagation algorithm is used to allocate and adjust the weight of each layer.(4)Improve the Faster RCNN feature extraction network,add BN layer to the feature extraction network,conduct batch normalization processing on the output of the feature extraction network,and verify the performance of its network model on INRIA data set.The improved pedestrian detection method in this thesis achieves an average accuracy of96% on INRIA dataset.Experimental verification shows that the algorithm in this thesis performs better in terms of speed and accuracy.