The Models Of Lane Detection And Semantic Segmentation Applied To Autonomous Driving

The development of self-driving cars has set off a revolution in automotive industry.Nowadays,traditional car companies,internet companies and an increasing number of start-up companies respectively do in-depth research about the techniques of autonomous driving.In addition,these techniques have been promoted to the national level,especially in China.The most important technique in automatic cars is scene perception and localization.Therefore,this paper mainly incorporates two fundamental parts which are lane detection and image semantic segmentation(SS).The textual content is as follow.A new lane detection method is proposed in this paper.First of all,an autonomous car obtains sequential colorful images.Then this method transforms these colorful images into corresponding gray images.The following step is that based on the camera parame-ters,we set the size of the map in Bird's Eye View(BEV)space and then transform above gray images into these maps in BEV space.What is more,this paper partitions a map into separated regions with the technique of Voronoi based on the previous calculated con-trol points.These independent regions are subsequently transformed into binary regions in which each pixel is 0 or 255.Moreover,according to the former lane line equation,we separate the image into different groups in which the coordinates of each non-zero pixel value will be recorded.Subsequently we apply the improved RANSAC algorithm to the grouped points to obtain the fitted lane line equation.Furthermore,the final lane line equation in BEV space will be calculated by incorporating the obtained lane line equation and the predicted control points using Kalman Filter.Finally,the final lane line equation will be transformed back into the perspective space.In conclusion,the proposed method can detect lane line as accurately as possible and it is robust for lane line detection.Moreover,this method can detect multiple lanes at the same time.Image semantic segmentation partitions an image into several coherent semantically meaningful parts,and classifies each part into one of the pre-determined classes.In this paper,we argue that existing SS methods cannot be reliably applied to autonomous driv-ing system as they ignore the different importance levels of distinct classes for safe-driving.For example,pedestrian,car,and bicyclist in the scene are much more important than sky and building when driving a car,so their segmentations should be as accurate as pos-sible.To incorporate the importance information possessed by various object classes,this paper designs an "Importance-Aware Loss"(IAL)that specifically emphasizes the criti-cal objects for autonomous driving.IAL operates under a hierarchical structure,and the classes with different importance are located in different levels so that they are assigned distinct weights.Furthermore,we derive the forward and backward propagation rules for IAL and apply them to four typical deep neural networks for realizing SS in intelligent driving system.The experiments on CamVid and Cityscapes datasets reveal that by employing the proposed loss function,the existing deep learning models including FCN,SegNet,ENet and ERFNet are able to consistently obtain the improved segmentation results on the pre-defined important classes for safe-driving.