基于红外与雷达的夜间无人车驾驶决策方法

速度与方向的决策建议是夜间无人车驾驶研究的关键,针对夜间无人车速度与方向决策,基于红外图像与雷达信息,提出了一种包含深度信息的红外图像多任务分类网络用来给出速度与方向决策.通过红外摄像头及雷达采集的数据训练深度网络,其中雷达采集的深度图像作为训练标签,采用卷积-反卷积神经网络来进行红外图像的深度估计,进而获得深度信息.利用深度信息制作分类网络训练标签,通过AlexNet分类网络得到速度决策建议.再根据红外图像的道路信息训练方向分类网络,将无人车的驾驶决策问题转化为分类模型,并将分类模型与深度估计网络相结合.实验结果表明,网络的角度准确率及速度准确率分别为87.43%和85.89%,并且利用训练得到的模型对图像进行决策的时间为0.04 s/帧,能够达到实时性的要求。

Nighttime unmanned vehicle driving decision method based on infrared and radar

The decision of speed and direction is the key to the study of night-time unmanned vehicle driving.Based on infrared image and radar information,a multi-tasking classification network of infrared images containing depth information is proposed for the decision of the speed and direction of unmanned vehicles at night.The depth network is trained by infrared cameras and data collected by radars.The depth-infrared image is estimated using a convolution-deconvolution neural network to obtain depth information.The depth information is used to make classification network training labels.The AlexNet classification network is used to obtain speed decision suggestions.Then according to the infrared image road information training direction classification network,the driving decision problem of the unmanned vehicle is transformed into a classification model,and the classification model is combined with the depth estimation network.Experimental results show that the angle accuracy and speed accuracy of the network are 87.43% and 85.89% respectively,and the time taken for decision making of the acquired image using the trained model is 0.04 s/frame,which can meet the requirements of real-time performance.