Search Results

Intelligent driving, aimed for collision avoidance and self-navigation, is mainly based on environmental sensing via radar, lidar and/or camera. While each of the sensors has its own unique pros and cons, camera is especially good at object detection, recognition and tracking. However, unpredictable environmental illumination can potentially cause misdetection or false detection. To investigate the influence of illumination conditions on detection algorithms, we reproduced various illumination intensities in a photo-realistic virtual world, which leverages recent progress in computer graphics, and verified vehicle detection effect there. In the virtual world, the environmental illumination is controlled precisely from low to high to simulate different illumination conditions in the driving scenarios (with relative luminous intensity from 0.01 to 400). Sedan cars with different colors are modelled in the virtual world and used for detection task.

Aimed to provide an effective solution for control-oriented applications, this paper proposes a novel method using a high-precision digital map to achieve high-accuracy positioning with fast updating rate. First, the map is developed using a high-definition LiDAR (Velodyne HDL 64E) and a RTK-GNSS system, which contains lane-level waypoints, road width, curb and typical obstacles along the road. Next, a robust version of ICP (Iterative Closest Point) is proposed to clean the corresponding points of large errors on map matching (MM). Finally, based on the large set of data from the environmental map, an unscented Kalman filter (UKF) is applied to fuse GNSS signal and dead reckoning (DR) to estimate the position. Thus the searching scope on the map can be considerably reduced so that the matching speed can be greatly improved. The high-precision digital map can be used not only for global path planning, but also for local driving detection and path planning.

LiDAR sensors have played more and more important role on Intelligent and Connected Vehicles (ICV) and Advanced Driver Assistance Systems (ADAS) .However, the development and testing of LiDAR sensors under real driving environment for ADAS applications are greatly limited by various factors, and often are impossible due to safety concerns. This paper proposed a novel functional LiDAR model under virtual driving environment to support development of LiDAR-based ADAS applications under early stage. Unlike traditional approaches on LiDAR sensor modeling, the proposed method includes both geometrical modeling approach and physical modeling approach. While geometric model mainly produces ideal scanning results based on computer graphics, the physical model further brings physical influences on top of the geometric model. The range detection is derived and optimized based on its physical detection and measurement mechanism.

Environmental sensing and perception is one of the key technologies on intelligent driving or autonomous vehicles. As a complementary part to current radar and lidar sensors, ultrasonic sensor has become more and more popular due to its high value to the cost. Different from other sensors mainly based on propagation of electromagnetic wave, ultrasonic sensor possesses some unique features and physical characteristics that bring many merits to autonomous vehicle research, like transparent obstacles and highly reflective surfaces detection. Its low-cost property can further bring down hardware cost to foster widespread use of intelligent driving or autonomous vehicles. To accelerate the development of autonomous vehicle, this paper proposes a high fidelity ultrasonic sensor model based on its physical characteristics, including obstacle detection, distance measurement and signal attenuation.

Vision-based Advanced Driver Assistance Systems has achieved rapid growth in recent years. Since vehicle field testing under various driving scenarios can be costly, tedious, unrepeatable, and often dangerous, simulation has thus become an effective means that reduces or partially replaces the conventional field testing in the early development stage. However, most of the commercial tools are lack of elaborate lens/sensor models for the vehicle mounted cameras. This paper presents the system-based camera modeling method integrated virtual environment for vision-based ADAS design, development and testing. We present how to simulate two types of cameras with virtual 3D models and graphic render: Pinhole camera and Fisheye camera. We also give out an application named Envelope based on pinhole camera model which refers to the coverage of Field-of-Views (FOVs) of one or more cameras projected to a specific plane.

This paper presents a novel approach of developing a vision-based forward collision warning system (FCW) under a virtual and real-time driving environment. The proposed environment mainly includes a 3D high-fidelity virtual driving environment developed with computer graphics technologies, a virtual camera model and a real-time hardware-in-the-loop (HIL) system with a driver simulator. Some preliminary simulation has been conducted to verify that the proposed virtual environment along with the image generated by a virtual camera model is valid with sufficient fidelity, and the real-time HIL development system with driver in the loop is effective in the early design, test and verification of the FCW and other similar ADAS systems.

Vision-based Advanced Driver Assistance Systems (Vi-ADAS) has achieved rapid growth in recent years. Since vehicle field testing under various driving scenarios can be costly, tedious, unrepeatable, and often dangerous, simulation has thus become an effective means that reduces or partially replaces the conventional field testing in the early development stage. This paper proposes a quantitative assessment framework for model quality evaluation of 3D scene under simulation platform. An imaging model is first built. The problem of solving the imaging model is then transformed into the problem of intrinsic image decomposition. Based on Retinex theory and Non-local texture analyses, a superior intrinsic image decomposition method is adopted to evaluate the fidelity of the 3D scene model through the degree of deviation to the Reflectance and Shading intrinsic maps respectively.