Research On Road Target Detection Based On Convolutional Neural Network

Target detection has always been a very important research topic in the field of computer vision.The main research is to detect and locate target objects in images from still pictures or videos.Road target detection is a research issue that finds out the road target in the scene image.With the development of unmanned driving,the detection of road targets shows great potential for development.During road driving,the vehicle's ability to sense surrounding objects can improve driving safety.Due to the complexity of the scene where the road target is located,there is no mature detection method for image-based road target detection,and more research is still needed in the practical application of the road target detection algorithm.In the traditional road target detection,it is necessary to manually extract the target features,and the generalization ability of the algorithm model is poor.Because convolutional neural network has the ability to automatically extract the target features of images,it can solve the problem of manually extracting the target features.Therefore,this paper uses the deep learning technology and proposes putting forward the deep convolutional neural networks to solve the problem of road targets.This article introduced the application background of the target detection and the development status at home and abroad,introduced the collection,preprocessing and division of the data set of this topic,and then introduced the basic knowledge related to the research of this topic,such as the historical background of the convolutional neural network.The basic principles of convolutional neural networks,as well as the role of different levels of structure in convolutional neural networks provide a theoretical basis for target detection algorithms.This paper uses convolutional neural network(SSD)model to solve the problem of road target detection.SSD model is divided into basic network and multi-dimension feature map detection.The basic network implements automatic extraction of image target features,and then extracts different sizes in the basic network.In the feature map,a convolution filtering is performed on the multi-dimensional feature map,and the target object's coordinate value and target category are finally obtained.In the experiment,the number of detection layers of the feature map was increased in the SSD model,the original image size was increased,modified basic network and the corresponding parameters were adjusted.After many iterations,the target model was finally obtained.The experiment adopts the image collected by the driving recorder.Three types of people are identified in the image: the vehicle,the pedestrian and the rider.The experiment shows that the smaller the detection target size is,the more difficult the detection is,and the worse the detection effect is.The SSD model detects the target.The average accuracy rate increased by 0.098.Compared with the traditional target recognition algorithm,the proposed road target detection method saves the manual feature extraction,reduces the workload,and improves the generalization ability of the model.