Research On Intelligent Vehicle Target Detection Based On Multi-sensor Information Fusion

Intelligent vehicles perceive the environment information around the vehicle body through the environment awareness system,which provides effective basic data for intelligent vehicle path planning and decision control.Therefore,the construction of environment awareness system is the most basic and key link to realize vehicle intelligence.In this paper,the object detection method of intelligent vehicle based on multi-sensor information fusion is studied.The specific research contents are as follows:(1)In order to solve the problem that the quality of images collected by visual imaging equipment is poor in complex environment such as haze,which affects the detection accuracy of visual objects.An improved AOD-Net neural network algorithm is proposed.Firstly,the mixing loss function is introduced to enhance the visual effect of the de-fogging image while improved the details of the de-fogging image.Then,the hybrid convolutional module is used to improved the receptive field of image feature extraction and enrich the feature information extraction ability.Finally,a comparative experiment of the proposed algorithm is carried out through the data set.Through the experimental analysis,the improved AOD-Net neural network algorithm has a good effect of removing fog.(2)An improved YOLOv5s target detection algorithm is proposed to solve the problem that the computing power of mobile terminal devices such as intelligent vehicles is limited.Firstly,the lightweight ShuffleNetv2 Block is used to redesign the backbone,which reduces the computation and parameters of the network model.Secondly,in the Neck part,the pyramid bidirectional weighted fusion module is adopted to improved the feature fusion between connections at different levels and enrich the semantic information.Then the GSConv module is used to reduce the complexity of the model.Then,the attention mechanism module is introduced in the Head part to improve the target detection performance of the model.Finally,the proposed improved YOLOv5s target detection algorithm is compared and simulated.The results show that the proposed target detection algorithm not only guarantees the detection accuracy,but also greatly reduces the model parameters,which is more conducive to the application of vehicle-mounted mobile devices.(3)Millimeter wave radar data extraction and effective target screening are completed.Firstly,according to the target detection requirements of the millimeter-wave radar on the vehicle end,the type of the millimeter-wave radar is determined,and the installation of the millimeter-wave radar on the vehicle end is completed.Then,data acquisition and analysis are completed according to the communication protocol of ARS404-21 millimeter wave radar.Finally,the targets in the radar analytic data are screened.The effective radar target information can be obtained by setting screening conditions.(4)An object detection framework based on multi-sensor information fusion is built.Firstly,according to the principle of multi-sensor fusion,the decision level fusion method is used to fuse two sensors.Secondly,the spatial and temporal matching of the visual and millimeter-wave radar was completed by using the coordinate conversion relationship between the two sensors and the least common multiple as the sampling period.Then,the GNN algorithm is used to correlate the data of the two sensors so that the data detects the same target.Finally,a test platform was built on the basis of intelligent vehicles,and a comparative analysis was carried out under two different working conditions,day and night.The experimental results show that the proposed scheme can effectively improve the stability and reliability of target detection and recognition,and reduce the problem of missing detection.