Design Of Crossing Monitoring Based On Radar Video Fusion

Video target detection is currently the focus of research and hotspot in the field of artificial intelligence.It plays a key role in medical diagnosis,autonomous driving,military reconnaissance,and other fields.Millimeter wave radar,as one of the common sensors in the field of transportation,can detect the distance,speed,angle and other parameters of the vehicle target,but the millimeter wave radar collection data lacks visibility and cannot get the measured target’s morphology and other detailed information;cameras,as the main road monitoring equipment,run in all corners of urban roads to monitor and manage various roads and vehicles,but the camera sensors are affected by light or rain and snow,and lack of depth information.Therefore,the traditional multi-target tracking technology is limited by the sensors themselves,which is difficult to meet the target tracking requirements in complex environments.The use of heterogeneous information obtained by different sensors can provide complete and effective information for target tracking,which helps to track the target with high accuracy and is an important development direction of multi-target tracking technology.In this paper,we introduce a data fusion method to fuse the target information obtained from millimeter wave radar and camera for the problem of low target alignment in complex environment and high signal-to-noise ratio scenarios,which is mainly as follows:First,the performance of millimeter wave radar is analyzed in terms of the detection range of the radar and the detection and differentiation ability of the targets.In terms of detection range,the transmitting power of the radar and the angle of the antenna are mainly analyzed;in terms of distinguishing ability,the three indicators of its speed resolution,distance resolution,and angle resolution are mainly considered.And the radar performance is improved based on MIMO sparse array antenna technology.Secondly,to address the problem of interference from factors such as image noise and complex background,the CA attention module is added to the Neck structure of YOLOv5 s,which is a new attention mechanism that can effectively improve the detection accuracy of small targets.The attention mechanism is used to update the feature map before fusion to ignore some irrelevant information and focus more attention on key features so that the feature map used for target detection after fusion contains more valid information.As well as proposing to replace the decoupling head,this can be used to improve the detection effect by using different perceptions also depending on the different sizes of the images.This decoupling head also allows the use of more convolutional layers,thus improving the representation of the network and thus better identifying small targets.Finally,the target data needs to be optimized before fusion.The first is to pair the time and space generated by different sensors one by one to ensure that environmental information with the same nature and characteristics can be generated after pairing for the fusion center to obtain later.Then,according to a certain process,the information collected by multi-source sensors is matched,and the data collected by different sensors on the same target is fused,which can produce more perfect target information.The reliability and accuracy of target detection can be greatly improved by real-time tracking and detection of fused targets.