Research On Typical Target Recognition Technology In Intelligent Assisted Driving Scenario

With the continuous development of technology and the improvement of people's living standards,more and more families have their own private cars and generally choose private cars for travel.With it comes problems such as road congestion,traffic accidents,and environmental pollution.In order to solve and alleviate the above problems,scholars and related persons from various countries are committed to the research of intelligent transportation systems.The research of "vehicle-pedestrian-lane line" detection technology,which is an important part of Advanced Assisted Driving System(ADAS),has achieved great practical significance.Recently,due to the significant increase in computing power of computer hardware and GPUs,deep learning has once again attracted people's attention,and object detection algorithms based on deep learning have emerged endlessly.Therefore,based on the theory of deep learning,this paper studies the detection of vehicles,pedestrians and lane lines in the scene of intelligent assisted driving.The main contents are as follows:Aiming at the real-time performance and hardware memory requirements in the intelligent driving assistance system,this paper presents an improved lightweight target detection model based on YOLOv3 and densely connected networks,which solves the problem that it cannot be transplanted on hardware due to the large number of model parameters.At the same time,the KL loss is used to solve the problem of inaccurate positioning of the prediction frame in YOLOv3.It achieves the sacrifice of a small amount of detection accuracy to effectively reduce the amount of model parameters,while ensuring the real-time performance of the model.Aiming at the problem of inaccurate model detection in lane line detection and detection failure in difficult scenes(such as at night or foggy weather),this paper proposes a method of lane line post-processing,which uses the knowledge of traditional image processing and the circular matrix kernel correlation tracking.The combination of algorithm(CSK)improves the problem of lane line detection failure in difficult scenarios and improves the accuracy of lane line detection.Finally,research on joint detection of vehicles and drivable areas is carried out.Utilizing the codec structure and multi-tasking idea,the features are extracted by sharing the convolutional neural network,and the branch network trains the vehicletarget detection and the driving area segmentation at the same time to realize the joint detection of the vehicle and the driving area.At the same time,three basic network structures were used for comparison experiments,which simplified the target detection network structure,effectively improved the model detection speed,and verified the validity of the model.The method proposed in this paper is verified by a large amount of data training and measured data sets.The results show that this method can achieve more efficient target detection in intelligent assisted driving applications.