| Intelligent driving and driverless cars are important development directions of vehicles.The traffic environment sensing technology of vehicles is indispensable,and the automatic detection and recognition of traffic lights is one of the important components of environmental perception.This technology is used in a wide range of applications,the automatic traffic lights recognition technology can provide assistant decisions for the color blindness and color weakness when they are acrossing the intersections.According to the different technical solutions adopted,the existing traffic light recognition schemes can be mainly divided into four categories based on vehicle road communication,based on surrounding vehicle state perception,GPS based navigation and vehicle vision based.Although multi-program integration is the trend of future development,traffic light recognition technology based on vehicle vision is still a hot topic in academic and business circles.The traditional traffic light detection algorithm mostly uses single-frame static images for detection and recognition,and the detection rate and missed detection rate need to be improved in complex traffic scenarios.To this end,this paper will introduce the time domain information of the image based on the optimization of traditional detection methods,focusing on solving the problems faced by traffic light detection and recognition tasks in complex traffic scenarios and high interference situations.The main work and research results of this paper are as follows:(1)The video image of the traffic light of the real vehicle is collected,and the traffic light of the 54-segment vehicle passing through the traffic light intersection is recorded.The traffic of the collection route is complicated and the interference is diverse.The collected weather includes daytime,nighttime and rainy days.The experimental data is representative,and the data is post-processed.Finally,the detection image database required for the research is constructed,and the algorithm training set and test set are identified.(2)This paper proposes a surround zero filtering algorithm to optimize the extraction of nighttime traffic lights.For the limitations of the backboard information,based on the scale distribution and the location distribution features further filter the pseudo traffic light area.(3)Based on the law of light and dark change of traffic lights,the algorithm of generating and updating the curve of light and dark of traffic lights is firstly proposed to describe the characteristics of light and dark changes of traffic lights in time windows.Then,a dynamic detection algorithm based on spatio-temporal joint is proposed.The effectiveness of the proposed algorithm is verified by comparison with traditional static feature-based detection algorithms.(4)In order to obtain the type of traffic light in the pending area and understand the instruction semantics,this paper first tests the traffic light recognition method using HOG feature and SVM classifier,and then proposes a traffic light recognition algorithm based on foreground horizontal and vertical projection.The recognition rate is comparable,but the latter is better in real time.Based on the recognition results,possible indication semantic is proposed.(5)Finally,this paper implements the above algorithm with C++ programming,completes the software interface development with qt5 platform,and finally completes the prototype of the system prototype based on the Linux platform and Raspberry Pi.Through experimental tests,the modules of the software and hardware system work normally,and the detection and recognition effects are stable. |
