We present a generalized and scalable method, called Gen-LaneNet, to detect 3D lanes from a single image. The method, inspired by the latest state-of-the-art 3D-LaneNet, is a unified framework solving image encoding, spatial transform of features and 3D lane prediction in a single network. However, we propose unique designs for Gen-LaneNet in two folds. First, we introduce a new geometry-guided lane anchor representation in a new coordinate frame and apply a specific geometric transformation to directly calculate real 3D lane points from the network output. We demonstrate that aligning the lane points with the underlying top-view features in the new coordinate frame is critical towards a generalized method in handling unfamiliar scenes. Second, we present a scalable two-stage framework that decouples the learning of image segmentation subnetwork and geometry encoding subnetwork. Compared to 3D-LaneNet, the proposed Gen-LaneNet drastically reduces the amount of 3D lane labels required to achieve a robust solution in real-world application. Moreover, we release a new synthetic dataset and its construction strategy to encourage the development and evaluation of 3D lane detection methods. In experiments, we conduct extensive ablation study to substantiate the proposed Gen-LaneNet significantly outperforms 3D-LaneNet in average precision(AP) and F-score.

Gen-LaneNet: A Generalized and Scalable Approach for 3D Lane Detection

Autonomous driving is experiencing rapid development. A lane-level map is essential for autonomous driving, and a lane-level road network is a fundamental part of a lane-level map. A large amount of research has been performed on lane-level road network generation based on various on-board systems. However, there is a lack of analysis and summaries with regards to previous work. This paper presents an overview of lane-level road network generation techniques for the lane-level maps of autonomous vehicles with on-board systems, including the representation and generation of lane-level road networks. First, sensors for lane-level road network data collection are discussed. Then, an overview of the lane-level road geometry extraction methods and mathematical modeling of a lane-level road network is presented. The methodologies, advantages, limitations, and summaries of the two parts are analyzed individually. Next, the classic logic formats of a lane-level road network are discussed. Finally, the survey summarizes the results of the review.

/pdf/lane-level-road-network-generation-techniques-for-lane-level-x90z6zt38x.pdf

Lane-level road network generation techniques for lane-level maps of autonomous vehicles: a survey

The vision-based road lane detection technique plays a key role in driver assistance system. While existing lane recognition algorithms demonstrated over 90% detection rate, the validation test was...

https://journals.sagepub.com/doi/pdf/10.1177/17298814211008752

Practical limitations of lane detection algorithm based on Hough transform in challenging scenarios

In order to solve the problem of the poor robustness for extracting the multi-lane marking, a multi-lane detection algorithm based on deep convolutional neural network is proposed. Since the results of the feature extraction of lane boundary were affected by various factors, a deep convolutional neural network based on FCN network is used for lane boundary feature extraction, and the neural network can be used to classify the lane images at the pixel level. Then the network parameters are trained on the public Tusimple data set and evaluated at Caltech Lanes data set. The last part combines a linear model with a curved model to realize the establishment of the lane marking equation, Hough transform is used to determine the fit interval and the least square method is used to fit lane marking. The experimental results show that the average accuracy of the algorithm for identifying lane on the Tusimple data set is 98.74%, and the accuracy rate on the Caltech Lanes data set is 96.29%.

/pdf/multi-lane-detection-based-on-deep-convolutional-neural-2ixmnmqiz3.pdf

Multi-Lane Detection Based on Deep Convolutional Neural Network

In this paper, we design a driver behavior monitoring and warning (DBMW) framework to detect dangerous driving for enhancing road safety through the Internet of Vehicles (IoV). The designed DBMW framework applies onboard image sensors and wearable devices to detect the deviation degree of vehicles and trace the head motion of drivers, respectively. According to our review of relevant research, DBMW is the first framework for driver behavior monitoring and warning that provides the following features: 1) DBMW can keep recognizing the located lane lines and estimating the power spectral density of lane deviation for a vehicle through image sensors, 2) DBMW can keep monitoring driver behaviors and measuring the anomaly level of a driver through wearable devices, and 3) DBMW can instantly send the warning messages of potential dangerous driving to neighboring vehicles and nearby pedestrians through IoV communications as necessary, which makes vehicles and pedestrians be aware of the existence of surrounding dangerous drivers in advance to keep alerting and avoid potential accidents/collisions. In particular, the prototype consisting of an Android-based sensing unit and an Arduino-based wearable device is implemented to verify the feasibility and superiority of DBMW. Experimental results show that DBMW outperforms existing methods and can significantly improve the detection accuracy and false alarm rates of dangerous driving behaviors.

Driver Behavior Monitoring and Warning With Dangerous Driving Detection Based on the Internet of Vehicles

Detection of road or lane is indispensable for the environmental perception of advanced driver assistance systems. It has been an active field of research with a wide application prospect. However, due to the complex illumination and interferences, such as vehicles and shadows in the real driving environment, lane detection is still a challenging task today. To address these issues, a robust method for road segmentation and lane detection based on a normal map is proposed. The first step of this approach is to generate the normal map by using the depth information and then extract a segmented road pavement without vehicles and buildings based on the normal map. Second, we improve an adaptive threshold segmentation method and denoising operations to enhance the lane markings. Third, the combination of Hough transform and vanishing point makes it more accurate to determine the starting points of host lanes, and then the lanes in the following image sequence can be detected in the adaptive region of interest. Compared with the state-of-the-art methods, the experimental results on the data sets in two countries demonstrate that our approach produces more credible performance under various light conditions or dense traffic.

Chang Yuan

Papers

Robust Lane Detection for Complicated Road Environment Based on Normal Map