自动驾驶车辆车道偏离识别及预警方法研究

Lane Departure Identification and Early Warning Method for Autonomous Vehicle

无人驾驶汽车需要通过视觉系统进行环境感知,而前方车道线识别与监测是行驶路线规划及行车安全性监测的重要部分。针对车道线图像识别实时性和适应性问题,提出了一种基于最大类间方差法和遗传算法相结合的车道线图像阈值分割优化算法。该算法通过二进制编码确定第1代种群,以最大类间方差计算公式作为适应度函数,并将类间方差作为适应度值,通过遗传算法计算出图像分割的阈值,利用MATLAB编程进行图像处理,在有效消除噪声的同时更好地保护了图像的细节,提高了车道线图像识别的准确度、适应度和识别速度;再根据车道线特征通过车道线检测跟踪算法进行拟合,利用车道偏离算法判断汽车在行驶过程中是否偏离车道线,将算法移植到了车载图像处理芯片,进行实车实验验证了算法的正确性;最后通过车载智能终端获取当前车道偏离的实时数据并发送至云服务器,以这些数据为基础判断车道偏离程度,并实现车道偏离警示。实验结果表明:所获取的车道偏离数据与当时实车行驶状态非常符合。

Driverless vehicles need to perceive the environment through the vision system,and lane recognition and monitoring at front is an important part of driving route planning and driving safety monitoring.Aiming at the problem of real-time and adaptability of lane-line image recognition,a lane-line image threshold segmentation optimization algorithm based on the combination of maximum inter-class variance method and genetic algorithmwas proposed.The proposed algorithm determined the first generation population by binary coding,took the maximum inter-class variance calculation formula as the fitness function,and used the variance between classes as the fitness value.The threshold of image segmentation was calculated by genetic algorithm and the image processing was performed by MATLAB programming,which effectively eliminated noise,protected the details of the image and improved the accuracy,fitness and recognition speed of lane-line image recognition.Then,according to the lane-line feature,the lane-line detection and tracking algorithm was used for fitting.The lane departure algorithm was used to judge whether the car deviated from the lane line during the driving process.The proposed algorithm was transplanted to the vehicle image processing chip,and the correctness of the proposed algorithm was verified by real vehicle experiments.Finally,the real-time data of the current lane departure was obtained by the vehicle intelligent terminal and sent to the cloud server.Based on these data,the degree of lane departure was judged and the lane departure warning was realized.The experimental results show that the obtained lane departure data is in good agreement with the actual vehicle driving state at that time.