Research On Vehicle Anti-collision Warning Method Based On Vision

In recent years,the development of Intelligent driving technology has been changing quickly.Intelligent driving technology has become a trend in the future.However,the current laws and regulations are not perfect,and the extreme complexity of actual traffic scenes determine that there is still a long way to go before fully automated driving.As the basic transition in intelligent driving,Advanced Driver Assistance System(ADAS)now has great market value and application space.The anti-collision warning function of the forward vehicle is an important basic function in ADAS,its core function is to detect and track the vehicle in front of the vehicle.This paper studies the vehicle detection and tracking algorithm and proves the feasibility of the method through experiments.Then,the monocular ranging function is completed for the need of the anti-collision warning method,build a prototype system that can run in real time on common hardware devices.The main work of this paper is as follows:(1)This paper proposed a vehicle cascade detector incorporating two feature operators.The detector takes full advantage of the prior knowledge of the vehicle’s distinctive edge features.First,in the primary classifier,the feature operator with fast calculation speed is used to quickly select the candidate region of the target vehicle,and then the second classifier is used to confirm the final vehicle position.The proposed method can improve the system detection efficiency while sacrificing the part without affecting the accuracy of the system function.(2)This paper proposed a vehicle detection and tracking framework based on kernelized correlation filters.The framework makes up for the shortcomings of the vehicle detection algorithm,and at the same time facilitates the management of the targets and the analysis of the target motion parameters.In view of the tracking drift caused by long-term tracking in the vehicle tracking process,an online monitoring strategy is introduced,and then an effective management mechanism for multiple targets is realized by combining actual traffic scenarios while tracking multiple targets simultaneously.In the multi-objective management module,the detector is triggered by a frame skipping method to discover the newly appearing target,and the corresponding supervisor is started between the plurality of target trackers by means of polling,and the multi-target tracking management mechanism ensures the Timely updates to the tracking queue while reducing time consumption.(3)Aiming at the visual ranging problem in the system,the calibration target system based on monocular vision is realized,which solves the problem of distance measurement.Combined with the vehicle detection and tracking module,a vehicle anti-collision warning prototype system based on visual ranging is designed.