Design Of Real-time Target Detection And Ranging System Based On Sensor Fusion

Environmental perception of intelligent agents represented by autonomous vehicles is the key technology to determine their level of intelligence,and is also the basis of active safety technology.In the context of the current environment perception dominated by the traditional single-sensor method,millimeter-wave radar can provide high precision and real-time distance ranging information.However,its disadvantages,such as high false alarm rate and low target positioning accuracy,are also obvious.Therefore,it has great limitations in the scene of intelligent environment perception and detection.In this case,multi-sensor fusion technology becomes the best choice to meet the requirements of complex environment perception and detection.Camera can provide more standard space information than radar,and at the same time more feasible target detection method.The addition of camera as a vision sensor for fusion can make up for the shortage of millimeter wave radar,at the same time can also cover up the shortcomings of camera ranging accuracy.Under the above background,this paper designs a target detection and ranging system based on sensor fusion.Firstly,the data transmission and acquisition of millimeter-wave radar in multi-sensor system are studied and an advanced clustering and filtering algorithm with forward-facing perspective is adopted to provide radar target data with high cleanliness and stability for the fusion system.Secondly,this paper studies the design of real-time video data processing module.The target detection model based on transfer learning is used for detection and tracking algorithm is used for image processing acceleration to output real-time image target data.Thirdly,the design of multi-sensor fusion system is studied.A complete sensor fusion model is constructed based on various complete theoretical systems,and the fusion decision method of MMW radar and camera target data is designed.Finally,through experimental verification,the system can detect and track the customized targets within the detection range selected by the user stably and accurately,so as to provide high-real-time multi-sensor fusion information.Compared with the traditional environmental perception scheme,this method improves the range and accuracy of environmental information collection and lays a foundation for the intelligent behavior subject’s information perception ability to adapt to the increasingly complex scene.