Research On Vehicle And Lane Line Detection Algorithm Based On Deep Learning

Vehicle and lane line detection are two important components in intelligent vehicle environment perception.Because of the excellent performance of deep learning in image recognition,vehicle and lane line detection based on deep learning has become a hot spot and focus of research.In this paper,vehicle and lane detection algorithms are studied based on in-depth learning,and the effectiveness of the algorithm is verified by experiments.The specific research contents are as follows:(1)Data sets used in vehicle and lane line detection models are established.From the collection of datasets to the labeling of datasets,the labeling is completed,then the corresponding format is converted,and finally the data set structure required by the model is organized by batch processing.(2)Design a vehicle detection algorithm based on deep learning.First,the image is downsampled to get different size feature maps.Then,based on the smallest feature maps,the feature maps with different sizes are sampled and stitched for multi-scale fusion.Then,three feature maps with different sizes are output for prediction of large-scale,medium-scale and small-scale targets.Before prediction,the K-means clustering algorithm is used to get three priori boxes of each feature map.Then,the location,confidence level and category of the vehicles are predicted by the cells in the feature map,and then non-maximum simulation is performed on all prediction boxes to detect the vehicles in the image.The results show that the algorithm can detect vehicles in different scenes,such as day,night,obstruction,etc.with high detection accuracy and good robustness.(3)Design a lane line detection algorithm based on deep learning.First,the image is downsampled by the encoder network to complete the feature extraction,then the image is segmented by the decoder network to complete the up-sampling,then the mean-shift algorithm is used for clustering post-processing to get the segmented results of the lane line pixel points,and then the least squares method is used for quadratic polynomial fitting of the lane line pixel points.Finally,the lane lines in the image are detected.The results show that the algorithm can detect lane lines well in different scenarios such as straight,curved,blurred and occlusion,and has a high detection accuracy and robustness.(4)The designed algorithm for vehicle and lane line detection is experimented.First,the software environment for vehicle and lane line detection network model is set up,and then the vehicle detection algorithm is validated on the self-collected vehicle dataset.The accuracy rate is 81.66%,the recall rate is 94.58%,and the F1 value is 87.65%.The lane line detection algorithm is validated on the Tucson dataset.The accuracy rate is 94.75%,the error rate is 8.11%,and the miss rate is 2.95%.The results show that the algorithm designed in this paper can accurately detect vehicles and lane lines in different scenarios,has good robustness,and provides a solution for vehicle and lane line detection in intelligent vehicle environment awareness.