Font Size: a A A

Research On Key Technologies Of Vehicle Blind Spot Detection And Early Warning Based On Multi-source Information Fusion

Posted on:2022-11-01Degree:MasterType:Thesis
Country:ChinaCandidate:K K RenFull Text:PDF
GTID:2492306764474134Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
In view of the current situation of frequent traffic accidents,it has become a trend to apply artificial intelligence technology to vehicle safety research.In order to improve the safety of vehicles on the road,thesis proposes a blind spot detection and lane change warning system based on multi-source information fusion.The current target detection model is large and cannot meet the real-time requirements on embedded devices,and the accuracy and reliability of the target information obtained by the monocular sensor is slightly insufficient.According to the above problems,thesis mainly does the following work:(1)First,the YOLOv4-Tiny detection model is optimized,and the model is compressed by replacing the ordinary convolution of the original model with a depthwise separable convolution.Then,a feature fusion module and a newly designed SPP module are added to the model to improve the model’s ability to extract features,so as to enhance the model’s detection performance for small targets such as pedestrians and cyclists.Through experimental comparison,the optimized model in thesis has a m AP of 72.77%,which is 1.36 points higher than the original model YOLOv4-Tiny.The weight size is only 5.1M,and the running speed on the Zynq7020 platform reaches22.6FPS.(2)In order to make up for the lack of accuracy and reliability of the target information obtained by the camera,the strategy of multi-source information fusion is used to increase the redundancy of information.According to the sensor installation position in thesis,a multi-sensor fusion scheme is proposed.The image target is regarded as the global object,and Kalman filter is used to track it.The image target and radar target are associated with the global object through the Hungarian algorithm.After the experimental test,the error rate of the fusion ranging within 50 meters does not exceed 3.4%.Compared with the monocular sensor,this scheme significantly improves the accuracy of target information.(3)According to the analysis of the safety distance of the driver’s lane-changing process and the reaction time of the driver and the car’s braking,a safe-distance model of the lane-changing process is designed,and a reasonable early warning strategy is formulated.The software and hardware of the early warning system are designed.After the test and analysis of the early warning system,it is verified that the lane change early warning system in thesis has a good early warning effect on the road.
Keywords/Search Tags:blind spot detection, lane change warning, multi-source information fusion
PDF Full Text Request
Related items