Erke Shang

Real-time lane departure warning system based on a single FPGA

AbstractThis paper presents a camera-based lane departure warning system implemented on a field programmable gate array (FPGA) device. The system is used as a driver assistance system, which effectively prevents accidents given that it is endowed with the advantages of FPGA technology, including high performance for digital image processing applications, compactness, and low cost. The main contributions of this work are threefold. (1) An improved vanishing point-based steerable filter is introduced and implemented on an FPGA device. Using the vanishing point to guide the orientation at each pixel, this algorithm works well in complex environments. (2) An improved vanishing point-based parallel Hough transform is proposed. Unlike the traditional Hough transform, our improved version moves the coordinate origin to the estimated vanishing point to reduce storage requirements and enhance detection capability. (3) A prototype based on the FPGA is developed. With improvements in the vanishing point-based steerable filter and vanishing point-based parallel Hough transform, the prototype can be used in complex weather and lighting conditions. Experiments conducted on an evaluation platform and on actual roads illustrate the effective performance of the proposed system.

Robust Unstructured Road Detection: The Importance of Contextual Information

Unstructured road detection is a key step in an unmanned guided vehicle (UGV) system for road following. However, current vision-based unstructured road detection algorithms are usually affected by continuously changing backgrounds, different road types (shape, colour), variable lighting conditions and weather conditions. Therefore, a confidence map of road distribution, one of contextual information cues, is theoretically analysed and experimentally generated to help detect unstructured roads. Two traditional algorithms, support vector machine (SVM) and k-nearest neighbour (KNN), are carried out to verify the helpfulness of the proposed confidence map. Following this, a novel algorithm, which combines SVM, KNN and the confidence map under a Bayesian framework, is proposed to improve the overall performance of the unstructured road detections. The proposed algorithm has been evaluated using different types of unstructured roads and the experimental results show its effectiveness.

Lane Detection Using Steerable Filters and FPGA-based Implementation

Vision-based lane detection is a key component for Driver-Assistance (DA) systems. It is still a challenging task in road scenes with complex shadows. This paper presents a novel local edge detector, using vanishing point position as a high level information to guide the use of steerable flters in lane detection, and its implementation on a Field Programmable Gate Array (FPGA) device. The FPGA technology has the advantages of high-performances for digital image processing and low cost, both of which are the requirements of DA systems. The main contributions of this work are twofold: 1) an edge extraction algorithm for lane detection is proposed, using the estimated vanishing point as high-level information to detect lanes. Firstly, a rough estimation of the vanishing point is used for calculating the expected local edge orientations. Secondly, a steerable flter is tuned to the expected direction for edge response. 2) a framework on FPGA is designed to implement the proposed algorithm. The framework is designed by using multi-engine technology, so it works in parallel for any order of steerable flters. Experiments and comparisons show that the proposed algorithm is very effcient in dealing with the complex shadow conditions, and works in real-time on FPGA device.

A real-time lane departure warning system based on FPGA

This paper presents a vision based Lane Departure Warning System (LDWS) and its implementation on a Field Programmable Gate Array (FPGA) device. It is used as a Driver Assistance (DA) system that supports drivers and helps avoiding accidents. The FPGA technology has the advantages of high-performances for digital image processing and low cost, both of which are the requirements of the DA systems. The main contributions of this work are threefold: 1) a hardware architecture, which combines Single Instruction Multiple Data (SIMD) structure and Single Instruction Single Data (SISD) structure based on FPGA, is implemented. This architecture is in possession of both efficiency and flexibility. Therefore, it can be employed to handle many vision processing tasks in real time; 2) an improved parallel Hough Transform (HT) is introduced. Compared with traditional HT, we move the origin to the estimated vanishing point, so as to reduce the storage requirements and improve the detection robustness; and 3) a simple and efficient warning strategy is presented which can be implemented on FPGA easily. Experiments illustrate the high performance of the introduced system in various common roadway scenes.

An Efficient Calibration Approach for Arbitrary Equipped 3-D LiDAR Based on an Orthogonal Normal Vector Pair

Light Detection And Ranging (LiDAR) has been widely employed in Unmanned Ground Vehicle (UGV) for autonomous navigation and object detection. In this paper, an efficient extrinsic parameter calibration approach, which is based on a pair of orthogonal normal vectors, is presented for an arbitrary equipped 3-D LiDAR. With the proposed approach, the whole calibration process can be easily and efficiently implemented in outdoor urban environment and no calibration equipment is required. The main advantages of this approach are twofold: (1) compared with traditional ways, the proposed approach employs an orthogonal normal vector pair, which is generated by ground plane and vertical wall in urban environment, so calibration equipments are not required anymore; (2) the normal vector is estimated from the point cloud data on a surface, thus a quite robust and accuracy estimation can be obtained. Experiments illustrate the effective and efficient performance of the proposed approach, compared with the state of the art.

A novel setup method of 3D LIDAR for negative obstacle detection in field environment

Negative obstacle detection is an important task for Unmanned Ground Vehicle (UGV) driving safely in field environments. This paper presents a novel 3D LiDAR setup method to deal with this issue. The proposed setup method has two advantages: 1) the blind area near the vehicle is greatly shrunken, which is very important in driving on narrow roads or taking a turning for the field UGV. 2) Compared to the traditional uprightly mounted LiDAR, the density of LiDAR data with this novel setup method is greatly improved, which is very useful both for positive and negative obstacle detection. With this new setup, a geometrical character based approach is introduced for the negative obstacle detection. Two cues, the width and the back of the negative obstacle are taken into consideration in this paper. Support Vector Machine (SVM) is employed to classify negative obstacles from the background. Meanwhile, these features are combined under a Bayesian framework. Experimental results show that the proposed setup method is useful and the proposed negative obstacle detection approach is effective.

A Path Planning Method for Assistant Parallel Car-Parking

Path planning is one of the key technologies in an intelligent car-parking system. This paper presents a path planning method for parallel parking mode. a scheme of a two-arc parking path is proposed, with the start position given but the goal to be yielded. the parking process is formulated as a constrained nonlinear optimization problem, where two objective functions are studied. and we gained the optimized results by MATLAB Optimization Toolbox.

OffRoadScene: An Open Database for Unstructured Road Detection Algorithms

We address the problem of unstructured road detection. This paper tries to build a database named OffRoadScene, which addresses the need for experimental data to quantitatively evaluate the performance of different unstructured road detection algorithms. OffRoadScene is comprised of two level of databases. In the first level, each frame document consists of not only image information, but also information of GPS (Global Position System), IMU (Inertial Measurement Unit) and laser scanner. In the second level, original images and corresponding benchmarks are offered. 550 series unstructured road images and 120 various scenarios of images are included currently. In addition, 13 video segments are in video segments file. In support of expanding OffRoadScene, we present a custom-made labeling software for assisting users who wish to add their own images. Finally, we explain how to use this database by evaluating some state-of-the-art unstructured road detection algorithms.

Region-of-interest generation for lane detection using high-level information

Vision-based lane detection is still a challenging task in real application with a variety of complex road scenes. What makes the problem even more difficult is the requirements of real time implement with on-board processor. To speedup the lane detection processing, the Region-of-Interesting (ROI) method was widely used to narrow the candidate searching range for local road features. In this paper, a continuous form of the ROI is first deduced from the lane model and the perspective projection. Then the factors that contribute to the appearance of the ROI are discussed. Lastly, an approximated form of the ROI is given for fast computing. In the proposed model, ROI is considered as a probability map, in which the value is corresponding to the possibility of the lane-markings to appear in a certain location in the scene. Moreover, the model takes the running manner of the vehicle and vibration of the vehicle into account. Experiments show that the proposed algorithm is very efficient in removing the outliers of local features in complex road scenes.

Lane detection using inversion transform

Vision based lane detection is an essential task in both autonomous land vehicles research and active safety system development. The straight lines in the image are always corresponding to the lanes to be detected, especially in the near front of the vehicle. Thus Hough transform is employed to extract lanes form the edge map of the road scenes, plenty of linear structure staffs, however, may appear in the scene, therefore spurious peaks in the Hough space may exist which will lead to a incorrect detection. In this paper, we proposes a inversion transform based lane detection method, which first maps those distractive lines in the image into circles, while remains the linearity of the lanes to be extracted; then an improved parallel Hough transform is followed to detect these lanes. Experimental results illustrates that the presented method are both efficient and robust in lane detection.