基于视频的移动车辆实时测距研究

A video-based research on real-time distance measurement of mobile vehicles

智能车辆的辅助驾驶系统对于周围环境的物体感知需快速且准确,在路况复杂、车速高的高速公路上要求则更加严格.传统的目标检测方法获取周围车况信息较少,且对于实时性检测有所欠缺.本文建立一种新型基于MobileNet-V2的SSD移动车辆实时检测与测距模型,利用深度可分离卷积结构在保持精度的同时减少网络参数的计算复杂度,加快检测速度;利用具有线性瓶颈的倒置残差结构保存更多的卷积特征,使得网络学习目标的深层次特征;用扩充的Pascal VOC2007数据集进行训练和评估,测试检测模型在复杂环境下的鲁棒性.利用实际交通场景的视频数据进行验证,结果表明:车辆检测网络达到83.74%@0.5IOU的检测精度,在NVIDIA GTX1050Ti上对视频的检测速度达到25FPS,在满足精度的同时提高检测实时性.对车辆检测信息进行处理,采用数据回归建模的方法判断本车安全行驶范围,在车速为0-120 km/h的范围内,测距误差基本保持在5%以内,为驾驶员安全行驶提供决策依据.

The intelligent vehicle’s assisted driving system needs to be fast and accurate for the perception of the surrounding environment.It is more demanding on highways with complex road conditions and high speeds.The traditional object detection method obtains less information about the surrounding conditions,and exists shortcomings of real-time detection.In this paper,a new mobile vehicle real-time detection and distance measurement model based on SSD MobileNet-V2 is established.The depth separable convolution structure is used to reduce the computational complexity of network parameters while maintaining accuracy,which can speed up the detection.The inverted residual structure is employed with linear bottlenecks to preserve more convolution features which make the network learn the deep features of the object.The extended Pascal VOC2007 data set is applied for training and evaluation to test the robustness of the detection model in complex environments.Test with video data from actual traffic scenarios,the resultsshow that the vehicle detection network achieves the detection accuracy of 83.74%@0.5 IOU,and the detection speed of the video on the NVIDIA GTX1050 Ti is 25 FPS,which achieves high real-time detection while satisfying the accuracy.At the same time,the vehicle detection information is processed,and the data regression model is used to determine the safe driving range of the vehicle.In the case of 0-120 km/h,the distance measurement error is kept within 5%,which provides a decision basis for the driver’s safe driving.