The Research On Road Object Detection And Its Android APP Development Based On Deep Learning

With the development of the times and the advancement of science and technology,in recent years,autonomous driving technology has become the direction of extensive research in the world,especially the rapid development of deep learning technology in recent years.The combination of deep learning and computer vision makes the detection and recognition with vision up to a high level.The auxiliary driving is an indispensable research topic before fully implementing autonomous driving.Road object detection and recognition is an important part of autonomous driving and assisted driving.The detection and recognization of road objects in driving video is studied in this thesis,and Android applications for the purpose of such detection and recognition is implemented on the Android platform.The main work of this thesis includes:(1)A road object detection data set is established.This thesis is aimed at the detection and recognization of the urban and suburban road objects in the driving video recorder.It is necessary to establish a data set of specific scenes.The models trained directly using the public dataset library are lower in accuracy and certain types of objects are not included in the public dataset library.Therefore,this thesis annotates and establishes a road object dataset in the driving video recorder.The dataset contains a total of 4 categories,with a total of 4000 images,of which 3000 are the training datasets,1000 the test datasets.(2)The image in the driving recorder is caused by motion blur during the driving process of the car.Before using the dataset to train the model,the image needs to be deblurred.This thesis uses the deblur algorithm based on DeblurGAN.The motion blur can be effectively removed.Compared with the traditional deblurring algorithm,the algorithm has the advantage that it does not need to estimate the fuzzy function,and the deblurring process is fast.(3)SSD detection algorithm is to detect the object of the road,which combines the advantages of the Faster-RNN algorithm and the YOLO algorithm.It is an outstanding algorithm in the two indexes of balance speed and accuracy,and is easy to transplant to themobile terminal,and the convolutional neural network ShuffleNet is used to replace the VGG16 network in the original SSD.(4)In order to test the system on platform,the corresponding Android application software is developed,realizing the real-time detection and recognition of road objects in the video recorder and the detection and recognization of single images.In the application development,we use C++ and Java hybrid programming techniques.