We present a reinforcement learning approach using Deep Q-Networks to steer a vehicle in a 3D physics simulation. Relying solely on camera image input the approach directly learns steering the vehicle in an end-to-end manner. The system is able to learn human driving behavior without the need of any labeled training data. An action-based reward function is proposed, which is motivated by a potential use in real world reinforcement learning scenarios. Compared to a naive distance-based reward function, it improves the overall driving behavior of the vehicle agent. The agent is even able to reach comparable to human driving performance on a previously unseen track in our simulation environment.

Learning how to drive in a real world simulation with deep Q-Networks

We study the problem of distilling knowledge from a large deep teacher network to a much smaller student network for the task of road marking segmentation. In this work, we explore a novel knowledge distillation (KD) approach that can transfer ‘knowledge’ on scene structure more effectively from a teacher to a student model. Our method is known as Inter-Region Affinity KD (IntRA-KD). It decomposes a given road scene image into different regions and represents each region as a node in a graph. An inter-region affinity graph is then formed by establishing pairwise relationships between nodes based on their similarity in feature distribution. To learn structural knowledge from the teacher network, the student is required to match the graph generated by the teacher. The proposed method shows promising results on three large-scale road marking segmentation benchmarks, i.e., ApolloScape, CULane and LLAMAS, by taking various lightweight models as students and ResNet-101 as the teacher. IntRA-KD consistently brings higher performance gains on all lightweight models, compared to previous distillation methods. Our code is available at https://github.com/ cardwing/Codes-for-IntRA-KD.

/pdf/inter-region-affinity-distillation-for-road-marking-1p77o35v73.pdf

Inter-Region Affinity Distillation for Road Marking Segmentation

Autonomous driving has become a popular research project. How to control vehicle speed is a core problem in autonomous driving. Automatic decision-making approaches, such as reinforcement learning (RL), have been applied to control the vehicle speed. However, the popular Q-learning algorithm is unstable in some games in the Atari 2600 domain. In this paper, a reinforcement learning approach called Double Q-learning is used to control a vehicle’s speed based on the environment constructed by naturalistic driving data. Depending on the concept of the direct perception approach, we propose a new method called integrated perception approach to construct the environment. The input of the model is made up of high dimensional data including road information processed from the video data and the low dimensional data processed from the sensors. During experiment, compared with deep Q-learning algorithm, double deep Q-learning has improvements both in terms of value accuracy and policy quality. Our model’s score is 271.73% times that of deep Q-learning.

Human-like Autonomous Vehicle Speed Control by Deep Reinforcement Learning with Double Q-Learning

The knowledge about the placement and appearance of lane markings is a prerequisite for the creation of maps with high precision, necessary for autonomous driving, infrastructure monitoring, lanewise traffic management, and urban planning. Lane markings are one of the important components of such maps. Lane markings convey the rules of roads to drivers. While these rules are learned by humans, an autonomous driving vehicle should be taught to learn them to localize itself. Therefore, accurate and reliable lane-marking semantic segmentation in the imagery of roads and highways is needed to achieve such goals. We use airborne imagery that can capture a large area in a short period of time by introducing an aerial lane marking data set. In this paper, we propose a symmetric fully convolutional neural network enhanced by wavelet transform in order to automatically carry out lane-marking segmentation in aerial imagery. Due to a heavily unbalanced problem in terms of a number of lane-marking pixels compared with background pixels, we use a customized loss function as well as a new type of data augmentation step. We achieve a high accuracy in pixelwise localization of lane markings compared with the state-of-the-art methods without using the third-party information. In this paper, we introduce the first high-quality data set used within our experiments, which contains a broad range of situations and classes of lane markings representative of today’s transportation systems. This data set will be publicly available, and hence, it can be used as the benchmark data set for future algorithms within this domain.

/pdf/aerial-lanenet-lane-marking-semantic-segmentation-in-aerial-3qdol2baue.pdf

Aerial LaneNet: Lane-Marking Semantic Segmentation in Aerial Imagery Using Wavelet-Enhanced Cost-Sensitive Symmetric Fully Convolutional Neural Networks

Curve is the traffic accident-prone area in the traffic system of the structural road. How to effectively detect the lane-line and timely give the traffic information ahead for drivers is a difficult point for the assisted safe driving. The traditional lane detection technology is not very applicable in the curved road conditions. Thus, a curve detection algorithm which is based on straight-curve model is proposed in this paper and this method has good applicability for most curve road conditions. First, the method divides the road image into the region of interest and the road background region by analyzing the basic characteristics of the road image. The region of interest is further divided into the straight region and the curve region. At the same time, the straight-curve mathematical model is established. The mathematical equation of the straight model is obtained by using the improved Hough transform. The polynomial curve model is established according to the continuity of the road lane-line and the tangent relationship between the straight model and the curve model. Then, the parameters of the curve model equation are solved by the curve fitting method. Finally, the detection and identification of the straight and the curve are realized respectively and the road lane-line is reconstructed. Experiments show that this method can accurately identify the curve lane-line, provide effective traffic information, make early warning, and it also has certain universality.

Lane Detection of Curving Road for Structural Highway With Straight-Curve Model on Vision

In this paper a new lane marking detection algorithm in different road conditions for monocular vision was proposed. Traditional detection algorithms implement the same operation for different road conditions. It is difficult to simultaneously satisfy the requirements of timesaving and robustness in different road conditions. Our algorithm divides the road conditions into two classes. One class is for the clean road, and the other one is for the road with disturbances such as shadows, non-lane markings and vehicles. Our algorithm has its advantages in clean road while has a robust detection of lane markings in complex road. On the remapping image obtained from inverse perspective transformation, a search strategy is used to judge whether pixels belong to the same lane marking. When disturbances appear on the road, this paper uses probabilistic Hough transform to detect lines, and finds out the true lane markings by use of their geometrical features. The experimental results have shown the robustness and accuracy of our algorithm with respect to shadows, changing illumination and non-lane markings.

Robust lane marking detection under different road conditions

A new robust lane marking detection algorithm for monocular vision is proposed. It is designed for the urban roads with disturbances and with the weak lane markings. The primary contribution of the paper is that it supplies a robust adaptive method of image segmentation, which employs jointly prior knowledge, statistical information and the special geometrical features of lane markings in the bird's-eye view. This method can eliminate many disturbances while keep points of lane markings effectively. Road classification can help us extract more accurate and simple characteristics of lane markings, so the second contribution of the paper is that it uses the row information of image to classify road conditions into three kinds and uses different strategies to complete lane marking detection. The experimental results have shown the high performance of our algorithm in various road scenes.

Lane marking detection based on adaptive threshold segmentation and road classification

The invention discloses a lane line historical frame recognition method. The method comprises performing Gaussian filter preprocessing on acquired road images and performing adverse projection on the images to be converted into airscapes; performing binarization processing on the airscapes through an adaptive threshold binarization method and performing Hough transformation on the airscapes to extract lines; determining which lines in Hough transformation results are lane lines based on the line angle LX.theta, the line distance LX.rho, the line ticket number LX.V and the line initial point distance LX.S of each line LX, wherein X =1, 2, ......N and the space and positions of the lane lines obtained in the previous period. By the aid of the method, interferences of complex road conditions of characters, multiple interference lines, shadow shielding, lane line damage, stain coverage and the like are removed effectively, and the recognition rate and the stability of lane lines are improved greatly; the method can be widely applied to car safety auxiliary driving systems to assist drivers for the maintaining of traveling of cars in the lanes in the monotonous driving environment.

Yan Song

Papers

Robust lane marking detection under different road conditions

Lane marking detection based on adaptive threshold segmentation and road classification

Lane line historical frame recognition method