Research On Key Technologies Of Laen Departure Warning Based On Deep Learning

Lane departure warning system is an important functional module of safe safety-assisted system.The system continuously monitors the relative position of its own vehicle and lane boundary,carries out full-time monitoring and evaluation of potential lane departure events,and reminds the driver by flashing signal lights,buzzer,steering wheel seat vibration and other measures when the vehicle departs from the lane.Based on the analysis of the functional requirements of the system,this paper studies the key technologies of lane departure early warning based on deep learning,such as lane detection on the structured road ahead of the vehicle,prediction of vehicle lateral departure distance,and construction of lane departure early warning model.The main work of this paper is as follows:(1)This paper proposed a lane detection model based on the steerable convolutional neural network.A steerable convolutional neural network was constructed for the problem of insufficient detection accuracy of lane lines on the front structured road.In order to enhance the description ability of lane features on the structured road ahead and improve the accuracy of lane detection,the steerable convolution neural network was constructed by combining the directional controllable filter and convolution neural network to capture the area where the lane was located,and the lane line features in the area where the lane was located were fitted by the lane pixel gradient and the least square method,so as to complete the whole lane detection process.In order to avoid the interference of the lane independent part of the road image to the lane detection,the vanishing point of the lane is selected by using probability voting,and then the region of interest is selected.The experimental results show that compared with the Hough transform and the typical convolution neural network detection method,the model in this chapter achieved a more accurate lane detection.(2)The paper proposed a ?-greedy long-short-time memory neural network model to predict the lateral deviation distance of a vehicle.The ?-greedy strategy was proposed to improve the loss function of the long-term and short-term memory neural network,and the neural network was optimized by combining the stochastic gradient descent algorithm.Data loss processing and data standardization were performed on the input data of the network through the Expectation-Maximization algorithm and Z-score normalization,respectively,which reduced the computational cost of neural network training and improved the prediction accuracy of the neural network.The experimental results showed that compared with the linear prediction model and the traditional long-short-term memory neural network prediction model,the model had less prediction error and achieved a satisfactory compromise in speed and accuracy.(3)This paper constructed a lane departure warning model based on the departure area warning parameters.In order to accurately warn the driver before the unintentional lane departure event occurs,the departure area of the vehicle from the lane was calculated by using the vehicle's lateral deviation distance predicted above,and the departure area was used as a warning parameter.And judged the deviation index in theoretical research.Finally,the effectiveness of this departure warning model was verified by various departure conditions contained in the lane line video collected by the real vehicle.The experimental results showed that compared with the CCP lane departure warning model and the TLC lane departure warning model,the warning model proposed in this paper was timely and had a low false alarm rate.