LDNet: End-to-End Lane Marking Detection Approach Using a Dynamic Vision Sensor

Modern vehicles are equipped with various driver-assistance systems,\nincluding automatic lane keeping, which prevents unintended lane departures.\nTraditional lane detection methods incorporate handcrafted or deep\nlearning-based features followed by postprocessing techniques for lane\nextraction using frame-based RGB cameras. The utilization of frame-based RGB\ncameras for lane detection tasks is prone to illumination variations, sun\nglare, and motion blur, which limits the performance of lane detection methods.\nIncorporating an event camera for lane detection tasks in the perception stack\nof autonomous driving is one of the most promising solutions for mitigating\nchallenges encountered by frame-based RGB cameras. The main contribution of\nthis work is the design of the lane marking detection model, which employs the\ndynamic vision sensor. This paper explores the novel application of lane\nmarking detection using an event camera by designing a convolutional encoder\nfollowed by the attention-guided decoder. The spatial resolution of the encoded\nfeatures is retained by a dense atrous spatial pyramid pooling (ASPP) block.\nThe additive attention mechanism in the decoder improves performance for high\ndimensional input encoded features that promote lane localization and relieve\npostprocessing computation. The efficacy of the proposed work is evaluated\nusing the DVS dataset for lane extraction (DET). The experimental results show\na significant improvement of $5.54\\%$ and $5.03\\%$ in $F1$ scores in multiclass\nand binary-class lane marking detection tasks. Additionally, the intersection\nover union ($IoU$) scores of the proposed method surpass those of the\nbest-performing state-of-the-art method by $6.50\\%$ and $9.37\\%$ in multiclass\nand binary-class tasks, respectively.\n