Research And Verification On Some Key Technologies In Traffic Sign Detection Used In Embedded Vehicle-mounted Early Warning

Vehicle-mounted early warning system can provide some necessary safety warning information for driver to improve driving safety,which includes a number of subsystems,such as traffic signs recognition,overspeed warning,etc.And real-time traffic sign detection is an important fundamental technology in these subsystems.In this paper,we mainly focus on color-based traffic sign location algorithm and hybrid switching multithreading task scheduling strategy,which are two key technologies of the real-time traffic sign detection.Some strategies which are suitable for the limited performance embedded system are proposed.And we also set up an embedded prototype to verify the effectiveness of these methods in the actual road environment.In order to test the effect of the traffic sign detection algorithm,we established a China traffic signs dataset and made it public available for other researchers,which is currently the only Chinese public dataset in this area.At present,the mainstream traffic sign detection and localization methods are mainly based on the local features of color and geometric shape.In this framework,red and yellow segmentation methods,which is the most important color in traffic sign detection,are studied in depth.In this paper,the basic principle,implementation method and segmentation results of current used methods are studied and compared.The comparison results show that these methods have some limitations,therefore a hybrid color segmentation strategy is proposed,which is based on these existing methods.This strategy achieves accurate and efficient segmentation of red and yellow in traffic signs by using a combination of several linear classifiers.The segmentation result is superior to the commonly used methods and the speed of the algorithm is similarly the same as that of the simplest RGB threshold method,so that we can guarantee the real-time requirements is satisfied in the small vehicle-mounted embedded equipment with limited performance.Based on the color segmentation result,we use the classical Hough circle transform to detect and locate the red circular traffic sign and evaluate the effect of the algorithm on the data set.In this paper,by modeling and analyzing traffic sign detection and recognition problems,we propose that sampling interval time can be used as an index to quantitatively measure the real-time performance of such systems.In addition,an optimal ideal multithreading task scheduling algorithm is proposed,which can significantly reduce the sampling interval time to improve the real-time performance of the system.However,the ideal task scheduling algorithm cannot be realized in practice.Therefore,we propose a practical hybrid switching task scheduling strategy and dynamic update parameter estimation strategy.Numerical simulation based on control system model and the actual test on embedded prototype both show that the methods proposed in this paper is effective to optimize the sampling interval time distribution and can improve the system real-time performance.We also develop an algorithm verification platform based on Qt and an embedded prototype based on Intel Joule module,and verifies the validity of the methods in this paper.Finally,the integrated test is carried out in the real road environment.The test results show that the methods proposed in this paper can meet the system real-time requirements in small embedded devices,and the detection rate is relatively high when the weather and light conditions are good.However,the robustness of the algorithm still needs to be strengthened.