Design And Implementation Of Vehicle Detection And Identification System

Locating and identifying vehicles in video images is important for implementing intelligent traffic regulation.If you can measure the size and color of the vehicle,you can accurately match the data of each known model,and it has good application value in the fields of road supervision and suspect tracking.This paper plans to identify the design and implementation of the prototype system,and improve the errors such as contours and shadows in the recognition method.In this paper,the research status and main methods of vehicle identification are studied.On this basis,the functions of object detection and instance segmentation are used.The convolutional neural network performs vehicle identification,calculates and acquires the information of the vehicle in the image by multi-view geometry,and designs and implements a system that automatically detects and identifies the video,the image passing through the vehicle,and extracts the maun feature information.The main research work and results of this paper are as follows:(1)The development history and research status of object detection,tracking,instance segmentation,multi-view geometry are studied.The open source software framework is selected according to design requirements and computing power to achieve the above objectives.Including the selection of the tensorflow deep learning framework to complete the recognition function,opencv to do video,image reading and processing,openmvg to achieve multi-view geometry to reconstruct the scene.(2)The core issues that this paper plans to solve are vehicle identification,instance segmentation,and size measurement.This paper chooses to use the mask R-CNN network framework to realize the recognition function,and adjusts and improves the network through the migration learning training of the self-labeled data set for vehicle identification;using multi-view geometry,camera model,projection matrix and other knowledge to achieve Measurement of vehicle size(3)In this paper,based on open source linear algebra library,multi-view geometry computing function library,etc.,multi-view geometry method are used to realize multiple models to reconstruct the model,and enhance the inclusiveness of the calculation results for measurement errors,shadows and other errors in the image.