车辆信息识别下的测距模型建立方法研究

Research on ranging model establishment method based on vehicle information recognition

为了探测车辆间的相对距离,避免危险车辆由于驾驶不当所引发的交通问题,提出一种深度学习目标识别下的跟驰车辆相对距离测定方法,避免了雷达测距的短距离局限性以及车辆未知性的缺点。该方法采用车载单目摄像机对侧后方车辆进行拍摄,实际物体和成像点之间的转换通过三坐标转换完成,利用深度网络识别目标车辆,获得目标车辆位置以及类别信息,并建立合适的测距模型,得到检测车辆与摄像头之间的相对距离,利用帧差法预测被检测车辆的行驶速度。选择河南省鹤壁市107,342国道进行试验,该路段验证了测距模型的有效性,静态测距下75 m以内相对误差控制在4%,速度误差控制在5%,因此,在检测到危险车辆的情况下,测距模型可以实现相对距离的实时准确性检测。

In order to detect the relative distance between vehicles and avoid traffic accidents caused by dangerous vehicles due to improper driving,a following vehicle relative distance measuring method based on deep learning target recognition is proposed,which avoids the limitation of short distance radar ranging and the shortcoming that a vehicle cannot get the situation around.In the proposed method,the vehicle-mounted monocular camera is used to shoot the rear side vehicles,and the transformation between the actual object and the imaging point is achieved by the transformation of three coordinates.The deep network is used to identify the target vehicle,obtain the position and category information of the target vehicle,and establish a suitable ranging model to obtain the relative distance between the detected vehicle and the camera.The frame difference method is used to predict the driving speed of the detected vehicle.The national highway G107 and G342 within Hebi city in Henan province were selected for the test to verify the effectiveness of the ranging model.The relative error and the speed error of static ranging within 75 m are kept at 4%and 5%,respectively.Therefore,the ranging model can realize real-time accurate detection of relative distance.