Research Of Target Detection And Tracking Algorithm Based On Sensory Fusion Of Radar And Camera

Unmanned driving is an important direction in the development of today’s and tomorrow’s technology,and target detection and tracking is a key part of the development process of unmanned technology.One of the core points of this process is the research of environment perception module,which is the most direct counterpart of the use of sensors.Faced with the uncertainties of unmanned scenarios and the complexity of road conditions,the shortcomings and deficiencies of single sensors in the process of target detection and tracking are gradually exposed,and the data information collected by single sensors alone is insufficient to adapt to the actual traffic environment.Thus,to ensure that the vehicle can obtain real-time information during the driving process,to ensure that the vehicle decision-making system can make timely and accurate decisions under the changing traffic scenarios and complex road conditions,and to ensure the safety of intelligent vehicles and passengers,the study of multi-sensor information fusion strategy is crucial.The article presents a method based on a multi-sensor information fusion target detection and tracking strategy.In the first place,single-sensor data acquisition and effective target acquisition using millimeter-wave radar and camera,respectively;then,data association algorithms are then used to correlate valid data between single sensors based on a decision-level information fusion approach;finally,the fusion determination strategy provided in the article,which aims to establish a reliable data support for planning and decision making when an intelligent vehicle is in operation.The major studies are as follows:(1)Millimeter wave radar-based data acquisition and effective target acquisitionIn the first place,the characteristics of millimeter wave radar are elaborated and the millimeter wave radar model used in this study is selected by comparing different regimes of millimeter wave radar.Then,we analyze the principles of radar range,speed and angle measurement under triangular wave modulation,and analyze the target data detected by millimeter wave radar;finally,we study the target acquisition method of millimeter wave radar.In order to deal with the problem that the original data detection results are messy and scattered,the target clustering strategy is used to deal with the problem;for the problem of false targets and other interfering signals,the queue exclusion method is used to reject them,and the Kalman filter algorithm and the life cycle algorithm are used to complete the consistency check and tracking of the valid targets.(2)Camera-based data acquisition and effective target acquisition.This paper compared different target detection algorithms and decided to improve the YOLOv5 target detection model based on a comprehensive analysis.The Ghost V2 module is used to replace part of the C3 module of the model backbone network for model pruning.At the same time,the Transformer structure with global informationawareness is introduced,which has the ability of parallel computation to reduce the waste of computational resources to a large extent,and its self-attentive mechanism greatly enhances the expression ability of the model.The adaptive spatial feature fusion mechanism ASFF is introduced to cope with the problem of varying target scales in the road target detection process.The small target detection layer is introduced to achieve real-time detection of obscured targets and distant small targets.The BDD100 K dataset is used to train the model and generate the model file for experimental validation of our trained model.(3)Designing a target detection and tracking strategy based on the combined use of millimeter wave radar and cameraDifferent fusion methods of multi-sensors are studied and the fusion principle between multi-sensors is analyzed.Since the point cloud targets collected by millimeter wave radar are two-dimensional while the targets in practical application scenarios are three-dimensional,the spatial fusion of multi-sensors is achieved by converting different coordinate systems to each other,and by analyzing the sampling periods of millimeter wave radar and camera,it is decided to choose the lower frequency millimeter wave radar sampling period as the reference for uniform sampling so as to achieve the temporal fusion of multi-sensors.After completing the temporal and spatial fusion,the article describes a target detection and tracking algorithm based on decisionlevel information fusion for the combined use of millimeter wave radar and camera,which integrates and processes data from multiple sensors through data correlation to obtain more accurate and comprehensive information.(4)Experimental validation and results analysis of the proposed algorithm are carried out by collecting target data under a variety of scenariosAn experimental platform with integrated millimeter wave radar and camera is established.The main application scenarios and functions of millimeter wave radar are different depending on the installation location,so we need to judge the actual application scenarios when determining the installation location.Once the build is complete we calibrate the sensor and obtain the camera’s internal reference matrix;target data in different scenarios are collected by the experimental vehicle platform for experiments,and the practicability of the multi-sensor information fusion algorithm studied in this paper is proved by analyzing the goal detection and tracking data of single sensor and multi-sensor information fusion,which proves its application value in real-world environments.