An End to End Spatial Point Instance Network for Lane Detection of Autonomous Driving

Lane detection is a critical aspect of autonomous driving systems. However, existing studies have indicated that the effectiveness of lane detection in complex scenarios is often below satisfactory levels. To address this challenge, we propose an integration of the encode-decode structure from PINet (Point Instance model), with SCNN, which leverages a spatial CNN. Through a comparative analysis of our approach against other methods, we have observed a remarkable superiority of our newly introduced model, SPINet (Spatial Point Instance model), specifically in complex scenarios. The exceptional performance of SPINet in the field of lane detection holds significant potential for enhancing the safety of autonomous driving. In summary, the notable advancements achieved by SPINet can make a substantial contribution to the safety of autonomous driving systems.