scholarly journals Lane Line Detection Based on Object Feature Distillation

Electronics ◽  
2021 ◽  
Vol 10 (9) ◽  
pp. 1102
Author(s):  
Malik Haris ◽  
Adam Glowacz
Keyword(s):  
Network Inference ◽  
Autonomous Driving ◽  
Lane Detection ◽  
Prediction Ability ◽  
Network Training ◽  
Training Stage ◽  
Knowledge Distillation ◽  
Lane Line ◽  
Object Feature ◽  
Forward Calculation

In order to meet the real-time requirements of the autonomous driving system, the existing method directly up-samples the encoder’s output feature map to pixel-wise prediction, thus neglecting the importance of the decoder for the prediction of detail features. In order to solve this problem, this paper proposes a general lane detection framework based on object feature distillation. Firstly, a decoder with strong feature prediction ability is added to the network using direct up-sampling method. Then, in the network training stage, the prediction results generated by the decoder are regarded as soft targets through knowledge distillation technology, so that the directly up-samples branch can learn more detailed lane information and have a strong feature prediction ability for the decoder. Finally, in the stage of network inference, we only need to use the direct up-sampling branch instead of the forward calculation of the decoder, so compared with the existing model, it can improve the lane detection performance without additional cost. In order to verify the effectiveness of this framework, it is applied to many mainstream lane segmentation methods such as SCNN, DeepLabv1, ResNet, etc. Experimental results show that, under the condition of no additional complexity, the proposed method can obtain higher F1Measure on CuLane dataset.

scholarly journals Strong-Weak Feature Alignment for 3D Object Detection

Electronics ◽  
2021 ◽  
Vol 10 (10) ◽  
pp. 1205
Author(s):  
Zhiyu Wang ◽  
Li Wang ◽  
Bin Dai
Keyword(s):  
Object Detection ◽  
Point Clouds ◽  
Autonomous Driving ◽  
Feature Representation ◽  
Alignment Algorithm ◽  
3D Object ◽  
3D Point Clouds ◽  
Object Feature ◽  
3D Object Detection ◽  
Feature Alignment

Object detection in 3D point clouds is still a challenging task in autonomous driving. Due to the inherent occlusion and density changes of the point cloud, the data distribution of the same object will change dramatically. Especially, the incomplete data with sparsity or occlusion can not represent the complete characteristics of the object. In this paper, we proposed a novel strong–weak feature alignment algorithm between complete and incomplete objects for 3D object detection, which explores the correlations within the data. It is an end-to-end adaptive network that does not require additional data and can be easily applied to other object detection networks. Through a complete object feature extractor, we achieve a robust feature representation of the object. It serves as a guarding feature to help the incomplete object feature generator to generate effective features. The strong–weak feature alignment algorithm reduces the gap between different states of the same object and enhances the ability to represent the incomplete object. The proposed adaptation framework is validated on the KITTI object benchmark and gets about 6% improvement in detection average precision on 3D moderate difficulty compared to the basic model. The results show that our adaptation method improves the detection performance of incomplete 3D objects.

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

2021 ◽  
Vol 18 (2) ◽  
pp. 172988142110087
Author(s):  
Qiao Huang ◽  
Jinlong Liu
Keyword(s):  
Hough Transform ◽  
Detection Rate ◽  
Real Life ◽  
Weather Conditions ◽  
Autonomous Driving ◽  
Detection Algorithm ◽  
Lane Detection ◽  
Search Range ◽  
Range Restriction ◽  
Lighting Conditions

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 usually conducted on limited scenarios. Significant gaps still exist when applied in real-life autonomous driving. The goal of this article was to identify these gaps and to suggest research directions that can bridge them. The straight lane detection algorithm based on linear Hough transform (HT) was used in this study as an example to evaluate the possible perception issues under challenging scenarios, including various road types, different weather conditions and shades, changed lighting conditions, and so on. The study found that the HT-based algorithm presented an acceptable detection rate in simple backgrounds, such as driving on a highway or conditions showing distinguishable contrast between lane boundaries and their surroundings. However, it failed to recognize road dividing lines under varied lighting conditions. The failure was attributed to the binarization process failing to extract lane features before detections. In addition, the existing HT-based algorithm would be interfered by lane-like interferences, such as guardrails, railways, bikeways, utility poles, pedestrian sidewalks, buildings and so on. Overall, all these findings support the need for further improvements of current road lane detection algorithms to be robust against interference and illumination variations. Moreover, the widely used algorithm has the potential to raise the lane boundary detection rate if an appropriate search range restriction and illumination classification process is added.

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

2021 ◽  
Vol 11 (22) ◽  
pp. 10713
Author(s):  
Dong-Gyu Lee
Keyword(s):  
Real Time ◽  
Computational Efficiency ◽  
Autonomous Driving ◽  
Learning Approaches ◽  
Practical Applications ◽  
Task Learning ◽  
Lane Line ◽  
High Level ◽  
Time Autonomous

Autonomous driving is a safety-critical application that requires a high-level understanding of computer vision with real-time inference. In this study, we focus on the computational efficiency of an important factor by improving the running time and performing multiple tasks simultaneously for practical applications. We propose a fast and accurate multi-task learning-based architecture for joint segmentation of drivable area, lane line, and classification of the scene. An encoder-decoder architecture efficiently handles input frames through shared representation. A comprehensive understanding of the driving environment is improved by generalization and regularization from different tasks. The proposed method learns end-to-end through multi-task learning on a very challenging Berkeley Deep Drive dataset and shows its robustness for three tasks in autonomous driving. Experimental results show that the proposed method outperforms other multi-task learning approaches in both speed and accuracy. The computational efficiency of the method was over 93.81 fps at inference, enabling execution in real-time.

scholarly journals Training Convolutional Neural Networks with Multi-Size Images and Triplet Loss for Remote Sensing Scene Classification

Sensors ◽  
2020 ◽  
Vol 20 (4) ◽  
pp. 1188 ◽  
Author(s):  
Jianming Zhang ◽  
Chaoquan Lu ◽  
Jin Wang ◽  
Xiao-Guang Yue ◽  
Se-Jung Lim ◽  
...  
Keyword(s):  
Remote Sensing ◽  
Classification Accuracy ◽  
Model Parameters ◽  
Classification Algorithms ◽  
Scene Classification ◽  
Training Strategy ◽  
Network Training ◽  
Training Stage ◽  
Triplet Loss ◽  
And Training

Many remote sensing scene classification algorithms improve their classification accuracy by additional modules, which increases the parameters and computing overhead of the model at the inference stage. In this paper, we explore how to improve the classification accuracy of the model without adding modules at the inference stage. First, we propose a network training strategy of training with multi-size images. Then, we introduce more supervision information by triplet loss and design a branch for the triplet loss. In addition, dropout is introduced between the feature extractor and the classifier to avoid over-fitting. These modules only work at the training stage and will not bring about the increase in model parameters at the inference stage. We use Resnet18 as the baseline and add the three modules to the baseline. We perform experiments on three datasets: AID, NWPU-RESISC45, and OPTIMAL. Experimental results show that our model combined with the three modules is more competitive than many existing classification algorithms. In addition, ablation experiments on OPTIMAL show that dropout, triplet loss, and training with multi-size images improve the overall accuracy of the model on the test set by 0.53%, 0.38%, and 0.7%, respectively. The combination of the three modules improves the overall accuracy of the model by 1.61%. It can be seen that the three modules can improve the classification accuracy of the model without increasing model parameters at the inference stage, and training with multi-size images brings a greater gain in accuracy than the other two modules, but the combination of the three modules will be better.

scholarly journals Multistructure-Based Collaborative Online Distillation

Entropy ◽  
2019 ◽  
Vol 21 (4) ◽  
pp. 357
Author(s):  
Liang Gao ◽  
Xu Lan ◽  
Haibo Mi ◽  
Dawei Feng ◽  
Kele Xu ◽  
...  
Keyword(s):  
Performance Improvement ◽  
Network Performance ◽  
State Of The Art ◽  
Autonomous Driving ◽  
Supplementary Information ◽  
Training Method ◽  
Machine Learning Methods ◽  
Information Diversity ◽  
Knowledge Distillation ◽  
Memory Resources

Recently, deep learning has achieved state-of-the-art performance in more aspects than traditional shallow architecture-based machine-learning methods. However, in order to achieve higher accuracy, it is usually necessary to extend the network depth or ensemble the results of different neural networks. Increasing network depth or ensembling different networks increases the demand for memory resources and computing resources. This leads to difficulties in deploying depth-learning models in resource-constrained scenarios such as drones, mobile phones, and autonomous driving. Improving network performance without expanding the network scale has become a hot topic for research. In this paper, we propose a cross-architecture online-distillation approach to solve this problem by transmitting supplementary information on different networks. We use the ensemble method to aggregate networks of different structures, thus forming better teachers than traditional distillation methods. In addition, discontinuous distillation with progressively enhanced constraints is used to replace fixed distillation in order to reduce loss of information diversity in the distillation process. Our training method improves the distillation effect and achieves strong network-performance improvement. We used some popular models to validate the results. On the CIFAR100 dataset, AlexNet’s accuracy was improved by 5.94%, VGG by 2.88%, ResNet by 5.07%, and DenseNet by 1.28%. Extensive experiments were conducted to demonstrate the effectiveness of the proposed method. On the CIFAR10, CIFAR100, and ImageNet datasets, we observed significant improvements over traditional knowledge distillation.

scholarly journals Lane Endpoint Detection and Position Accuracy Evaluation for Sensor Fusion-Based Vehicle Localization on Highways

Sensors ◽  
2018 ◽  
Vol 18 (12) ◽  
pp. 4389 ◽  
Author(s):  
Eun Jang ◽  
Jae Suhr ◽  
Ho Jung
Keyword(s):  
High Speed ◽  
Autonomous Driving ◽  
Lane Detection ◽  
Accuracy Evaluation ◽  
Endpoint Detection ◽  
Driver Assistance Systems ◽  
Vehicle Localization ◽  
Position Accuracy ◽  
Essential Components ◽  
Mobile Mapping System

Landmark-based vehicle localization is a key component of both autonomous driving and advanced driver assistance systems (ADAS). Previously used landmarks in highways such as lane markings lack information on longitudinal positions. To address this problem, lane endpoints can be used as landmarks. This paper proposes two essential components when using lane endpoints as landmarks: lane endpoint detection and its accuracy evaluation. First, it proposes a method to efficiently detect lane endpoints using a monocular forward-looking camera, which is the most widely installed perception sensor. Lane endpoints are detected with a small amount of computation based on the following steps: lane detection, lane endpoint candidate generation, and lane endpoint candidate verification. Second, it proposes a method to reliably measure the position accuracy of the lane endpoints detected from images taken while the camera is moving at high speed. A camera is installed with a mobile mapping system (MMS) in a vehicle, and the position accuracy of the lane endpoints detected by the camera is measured by comparing their positions with ground truths obtained by the MMS. In the experiment, the proposed methods were evaluated and compared with previous methods based on a dataset acquired while driving on 80 km of highway in both daytime and nighttime.

scholarly journals Spatial Attention Fusion for Obstacle Detection Using MmWave Radar and Vision Sensor

Sensors ◽  
2020 ◽  
Vol 20 (4) ◽  
pp. 956 ◽  
Author(s):  
Shuo Chang ◽  
Yifan Zhang ◽  
Fan Zhang ◽  
Xiaotong Zhao ◽  
Sai Huang ◽  
...  
Keyword(s):  
Spatial Attention ◽  
Autonomous Driving ◽  
Obstacle Detection ◽  
Superior Performance ◽  
Generation Model ◽  
Fusion Method ◽  
Vision Sensor ◽  
Radar Images ◽  
Network Training ◽  
Safety Consideration

For autonomous driving, it is important to detect obstacles in all scales accurately for safety consideration. In this paper, we propose a new spatial attention fusion (SAF) method for obstacle detection using mmWave radar and vision sensor, where the sparsity of radar points are considered in the proposed SAF. The proposed fusion method can be embedded in the feature-extraction stage, which leverages the features of mmWave radar and vision sensor effectively. Based on the SAF, an attention weight matrix is generated to fuse the vision features, which is different from the concatenation fusion and element-wise add fusion. Moreover, the proposed SAF can be trained by an end-to-end manner incorporated with the recent deep learning object detection framework. In addition, we build a generation model, which converts radar points to radar images for neural network training. Numerical results suggest that the newly developed fusion method achieves superior performance in public benchmarking. In addition, the source code will be released in the GitHub.

Sharp Curve Lane Detection for Autonomous Driving

2019 ◽  
Vol 21 (2) ◽  
pp. 80-95 ◽  
Author(s):  
Hongzhe Liu ◽  
Xuewei Li
Keyword(s):  
Autonomous Driving ◽  
Lane Detection
Sign in / Sign up

Export Citation Format

Share Document