Enhanced free space detection in multiple lanes based on single CNN with scene identification

Abstract

Many systems for autonomous vehicles' navigation rely on lane detection. Traditional algorithms usually estimate only the position of the lanes on the road, but an autonomous control system may also need to know if a lane marking can be crossed or not, and what portion of space inside the lane is free from obstacles, to make safer control decisions. On the other hand, free space detection algorithms only detect navigable areas, without information about lanes. State-ofthe-art algorithms use CNNs for both tasks, with significant consumption of computing resources. We propose a novel approach that estimates the free space inside each lane, with a single CNN. Additionally, adding only a small requirement concerning GPU RAM, we infer the road type, that will be useful for path planning. To achieve this result, we train a multi-task CNN. Then, we further elaborate the output of the network, to extract polygons that can be effectively used in navigation control. Finally, we provide a computationally efficient implementation, based on ROS, that can be executed in real time. Our code and trained models are available online.

Deep Learning in Lane Marking Detection: A Survey

“…Although the softmax based cross entropy has demonstrated remarkable performance in various classification tasks, many studies still focus on improving its performance to tackle some long-existing problems such as hard sample mining and class imbalance, and these variants can also be found in lane marking detection. For example, [44] further regularizes the intra-class distance of the samples in the same class; [58], [75], [76] allocate different weights to different classes to alleviate training problems brought by unbalanced classes.…”

“…[100] first detects lane region mask and then calculates the convex hull surrounding the mask to attain the lane marking. A idea similar to [100] is used in [58].…”

“…In the downsampling block, the strategy of parallel pooling and convolution is also adopted. [58]- [60] use ERFNet as a backbone to improve the efficiency of lane marking detection.…”

Fast Drivable Areas Estimation with Multi-Task Learning for Real-Time Autonomous Driving Assistant

“…Multi-task deep learning has also been used for geometry and regression tasks. Fabio et al [2] detect navigable area by estimating free space inside each lane. The homoscedastic uncertainty estimation is used to achieve better performances.…”

“…The drivable area estimation for safe driving requires a more high-level understanding than other tasks such as free space detection, vehicle detection, pedestrian detection, and lane line detection [1]. According to the current environment in real-time, a drivable area is necessarily divided into ego-lanes, other drivable lanes, sidewalks, and so forth [2].…”