SOLOLaneNet: Instance Segmentation-Based Lane Detection Method using Locations

Modern cars are incorporating an increasing number of driver assist features, among which automatic lane keeping. The latter allows the car to properly position itself within the road lanes, which is also crucial for any subsequent lane departure or trajectory planning decision in fully autonomous cars. Traditional lane detection methods rely on a combination of highly-specialized, hand-crafted features and heuristics, usually followed by post-processing techniques, that are computationally expensive and prone to scalability due to road scene variations. More recent approaches leverage deep learning models, trained for pixel-wise lane segmentation, even when no markings are present in the image due to their big receptive field. Despite their advantages, these methods are limited to detecting a pre-defined, fixed number of lanes, e.g. ego-lanes, and can not cope with lane changes. In this paper, we go beyond the aforementioned limitations and propose to cast the lane detection problem as an instance segmentation problem - in which each lane forms its own instance - that can be trained end-to-end. To parametrize the segmented lane instances before fitting the lane, we further propose to apply a learned perspective transformation, conditioned on the image, in contrast to a fixed ”bird’s-eye view” transformation. By doing so, we ensure a lane fitting which is robust against road plane changes, unlike existing approaches that rely on a fixed, predefined transformation. In summary, we propose a fast lane detection algorithm, running at 50 fps, which can handle a variable number of lanes and cope with lane changes. We verify our method on the tuSimple dataset and achieve competitive results.

Aiming at the imbalance between the accuracy and real-time performance of lane detection, this paper proposes a novel vehicle road detection based on pyramid network and self-attention mechanism (namely PFSA). In this method, a two-sided multi-scale fusion network is used to realise the information exchange between shallow features and deep features, and to obtain contextual semantics. A new asymmetric convolution pyramid module is proposed, which fuses asymmetric convolution into cavity convolution layers with different expansion rates to improve feature extraction capability and reduce computation. A two-stage training method was used to train on a public data set and compared with other advanced lane detection methods. The experimental results show that the accuracy is 98.3% and the detection speed is 18.391 frames per second (fps), which is much better than other lane detection methods.

<div class="section abstract"><div class="htmlview paragraph">Lane detection plays a critical role in autonomous vehicles for safe and reliable navigation. Lane detection is traditionally accomplished using a camera sensor and computer vision processing. The downside of this traditional technique is that it can be computationally intensive when high quality images at a fast frame rate are used and has reliability issues from occlusion such as, glare, shadows, active road construction, and more. This study addresses these issues by exploring alternative methods for lane detection in specific scenarios caused from road construction-induced lane shift and sun glare. Specifically, a U-Net, a convolutional network used for image segmentation, camera-based lane detection method is compared with a radar-based approach using a new type of sensor previously unused in the autonomous vehicle space: radar retro-reflectors. This evaluation is performed using ground truth data, obtained by measuring the lane positions and transforming them into pixel coordinates. The performance of each method is assessed using the statistical <i>R</i><sup>2</sup> score, indicating the correlation between the detected lane lines and the ground truth. The results show that the U-Net camera-based method exhibits limitations in accurately detecting and aligning the lane lines, particularly in challenging scenarios. However, the radar-based lane detection method demonstrates a strong correlation with the ground truth which implies that the use of this sensor may improve current reliability issues from conventional camera lane detection approach. Furthermore, the study highlights the limitations of the U-Net model for camera lane detection, especially in scenarios with sun glare. This study shows that infrastructure-based radar retro-reflectors can improve autonomous vehicle lane detection reliability. The integration of different sensor modalities and the development of advanced computer vision algorithms are crucial for improving the accuracy, reliability, and energy efficiency of lane detection systems. Addressing these challenges contributes to the advancement of autonomous vehicles and the realization of safer and more efficient transportation systems.</div></div>

A robust and real-time lane detection method in low-light scenarios to advanced driver assistance systems

Lane detection plays a pivotal role in the successful implementation of Advanced Driver Assistance Systems (ADASs), which are essential for detecting the road’s lane markings and determining the vehicle’s position, thereby influencing subsequent decision making. However, current deep learning-based lane detection methods encounter challenges. Firstly, the on-board hardware limitations necessitate an exceptionally fast prediction speed for the lane detection method. Secondly, improvements are required for effective lane detection in complex scenarios. This paper addresses these issues by enhancing the row-anchor-based lane detection method. The Transformer encoder–decoder structure is leveraged as the row classification enhances the model’s capability to extract global features and detect lane lines in intricate environments. The Feature-aligned Pyramid Network (FaPN) structure serves as an auxiliary branch, complemented by a novel structural loss with expectation loss, further refining the method’s accuracy. The experimental results demonstrate our method’s commendable accuracy and real-time performance, achieving a rapid prediction speed of 129 FPS (the single prediction time of the model on RTX3080 is 15.72 ms) and a 96.16% accuracy on the Tusimple dataset—a 3.32% improvement compared to the baseline method.

In driving assistance systems, lane detection can provide significant information for driving safety. In this paper, we proposed a novel real-time lane detection method to extract the location of lane marking lines based on inverse perspective mapping transform (top view) for the region of interest (ROI) of a video frame. The data were then filtered by a selective oriented Gaussian high pass filter, Hough transformation, and Kalman filter to give the initial regions to our optimized RANSAC (Random Sample Consensus) Bezier splines fitting algorithm, which is the main innovation in this paper. Our experimental results and accuracy evaluation indicated that the proposed lane detection algorithm could run robustly in real time, and could achieve an average speed of 32.32 ms per frame for a 320 x 240 pixel image and 41.64 ms for a 640 x 480 pixel image, with a correct detection rate of over 92.85%. Moreover, our method was found to be suitable for various conditions.

Lane detection is crucial for driver assistance systems. However, road scenes are severely degraded in dense fog, which leads to the loss of robustness of many lane detection methods. For this problem, an end-to-end method combining polarimetric dehazing and lane detection is proposed in this paper. From images with dense fog captured by a vehicle-mounted monochrome polarization camera, the darkest and brightest images are synthesized. Then, the airlight degree of polarization is estimated from angle of polarization, and the airlight is optimized by guided filtering to facilitate lane detection. After dehazing, the lane detection is carried out by a Canny operator and Hough transform. Having helped achieve good lane detection results in dense fog, the proposed dehazing method is also adaptive and computationally efficient. In general, this paper provides a valuable reference for driving safety in dense fog.

Enhanced Cross Layer Refinement Network for robust lane detection across diverse lighting and road conditions

This research paper introduces an innovative method for real-time autonomous lane detection utilizing a Raspberry Pi embedded system and the OpenCV computer vision library. The primary objective is to improve vehicle navigation capabilities by precisely identifying lane boundaries from input video streams. The system's robust performance across diverse environmental conditions is achieved through the computational power of the Raspberry Pi platform and the flexibility of OpenCV. The methodology encompasses image preprocessing, lane detection algorithms, and seamless integration with the Raspberry Pi hardware. Through comprehensive experimental results, the proposed approach showcases its effectiveness and efficiency in autonomously detecting lanes. This research contributes to the development of safer and more reliable autonomous driving systems

Lane detection and tracking technique is the autonomous driving basis for Rubber-Tired Gantries (RTGs), vital to the automation and intelligence updating of man-driven container terminals. However, the existing lane detection methods developed for common road scenarios cannot meet the high-precision and robust all-weather requirements of RTG autonomous driving. In this article, we propose an Adaptive Edge-based Lane Detection and Tracking method considering RTG lanes’ characteristics in this paper. First, the candidate edges of lane lines are detected and paired based on the enhanced gradient features. Next, inverse perspective mapping is employed to search the right edges, followed by an adaptive sliding-window method. Ultimately, we develop an adaptive Kalman filter to track lane lines robustly, detecting confidence weighting by relaxing the constraint of lane line width. The proposed method is tested in an actual container yard, the lane centerline’s average position error is 2.051 pixels, and the detection success rate is close to 100%.

This paper introduces a robust yet straightforward lane detection and lateral control approach via the deployment of a dual camera based on the look-down strategy for autonomous vehicles. Unlike traditional single-camera systems that rely on the look-ahead methodology and a single front-facing preview, the proposed algorithm leverages two downward-facing cameras mounted beneath the vehicle’s driver and the passenger side mirror, respectively. This configuration captures the road surface, enabling precise detection of the lateral boundaries, particularly during lane changes and in narrow lanes. A Proportional-Integral-Derivative (PID) controller is designed to maintain the vehicle’s position in the center of the road. We compare this system’s accuracy, lateral steadiness, and computational efficiency against (1) a conventional bird’s-eye view lane detection method and (2) a popular deep learning-based lane detection framework. Experiments in the CARLA simulator under varying road geometries, lighting conditions, and lane marking qualities confirm that the proposed look-down system achieves superior real-time performance, comparable lane detection accuracy, and reduced computational overhead relative to both traditional bird’s-eye and advanced neural approaches. These findings underscore the practical benefits of a straightforward, explainable, and resource-efficient solution for robust autonomous vehicle lane-keeping.

Lane detection is the key technology of an advanced driver assistance system. This study proposed an efficient and novel lane detection method, Lane Kernel Parameter Module ResNet-18 (LKPM-ResNet-18), to improve the detection efficiency of lane with complex topologies (such as divergent lines and dense lines). This method can precisely predict the global structure in a bottom-up manner and directly output the curve parameters of the lane alignment model. In addition, LKPM was used to obtain accurately the local structure information and the slender shape of the lane with complex topology, and then the shape convolution kernel parameters of each instance are dynamically predicted to optimize the results of the subsequent output curves. Finally, the prior information of the global lane is fully integrated through the lane structure constraint loss, and the optimization results are fused into the global geometric curve fitting output, so that the fitted lanes are smoother and more continuous. The method proposed has real-time detection efficiency and requires little postprocessing because of the end-to-end direct approach. The validated results show that the efficiency of this method is obviously better than that of State-Of-The-Art (SOTA) methods, such as PointLaneNet and ENet-SAD. The F1 scores increase by 0.43% and 4.92%, respectively compared with the latest CurveLanes-NAS-L and SCNN on the CULane test dataset. The accuracy of 88.98% can be obtained on the CurveLanes test dataset. Meanwhile, the result from real driving environment detection exceeds the top performances of the other by 2.22% with the accuracy of 86.47%.

As the high growth of the number of vehicles, the traffic accidents are becoming more and more serious in recent years. In order to avoid the drivers being in danger, an intelligent vision-based system should focus on the image contents of the front the camera setting under the rear-view mirror in the vehicle. In this paper, we present a functional-link-based neuro-fuzzy network (FLNFN) structure for lane detection and departure warning system application. The proposed FLNFN model uses a functional link neural network (FLNN) to the consequent part of the fuzzy rules. This study uses orthogonal polynomials and linearly independent functions in a functional expansion of the FLNN. Thus, the consequent part of the proposed FLNFN model is a nonlinear combination of input variables. An online learning algorithm, which consists of structure learning and parameter learning, is also presented. The structure learning depends on the entropy measure to determine the number of fuzzy rules. The parameter learning, based on the gradient descent method, can adjust the shape of the membership function and the corresponding weights of the FLNN. The lane detection method and the departure warning system proposed in this paper have been successfully evaluated on a PC platform of 3.2-GHz CPU, where the average frame-rate is up to 30fps.

Dynamically Enhanced lane detection with multi-scale semantic feature fusion

Microscopic vehicle behaviors, such as car following, overtaking, and lane changing, may occur differently on links with different numbers of lanes. Thus, this paper addresses the impact of the number of lanes on the platoon dispersion of traffic flow in low-friction conditions. A well-adopted dispersion model is used to reflect the platoon dispersion. The platoon dispersion factor of the model is then recalibrated using data on road segments with different numbers of lanes. The data are obtained from a comprehensive survey taken at several locations on the arterial streets of Changchun and Hangzhou city, China, where each direction of road segment has two, three, four, or five lanes. The statistical results verify that the number of lanes does have an evident effect on the platoon dispersion factor α: when the number of lanes grows from two to five, the value of α decreases significantly; meanwhile, the slope of the function curve between α and the normalized flow becomes larger.