Unmanned electric buses are generally used to the ferry service. Their operating scenarios are closed or semi-enclosed parks whose road conditions are simple. In addition, this type of product has become the first choice for many autopilot start-up companies because of the small size, simple structure and short development cycle,.After investigation, it was found that the technology gap between domestic and foreign companies in unmanned electric buses is still relatively large. The unmanned electric buses of several foreign companies have already been put into operation through experiments.On their official website, detailed manuals of products and videos of actual operation are available, but these are not available on the domestic companies. It is noted that currently almost all of the driverless car perception systems use laser radars and cameras, indicating that the perceptual systems using these two sensors are both better at cost and perceptual effects.
The unmanned electric bus designed for this article has a driverless level of level 4, which runs on a structured road in the urban environment. Structured roads refer to standardized roads with clear lane markings and road boundaries. Due to the relative ease of urban transportation, good road conditions and less dust, the database is simple. Therefore, unmanned electric buses achieve the following functions:

(1) Road recognition in the case of lane lines

(2) edge detection

(3) Traffic Signs and Signal Detection

(4) Pedestrians and Vehicle Inspection

2. 研究的基本内容与方案

If the environment awareness system is modularized, it can be divided into three modules:

(1) Exterior environment perception module: All the above functions can be realized by using machine vision whose sensors are CMOS/CCD cameras; the laser radar can achieve functions more reliable. This article selects the laser radar and camera to detect simultaneously.

(2) Navigation and Positioning Module: There are currently four major directional positioning methods:

Scenario 1: Using GPS positioning, the attitude angles are obtained by differentiating the position sequence.

Advantages: simple system, low cost, easy to implement.

Disadvantages: Depends on GPS positioning accuracy. Can't locate orientation in GPS blind spot.

Option 2: Fusion GPS and inertial navigation system positioning and orientation.

Advantages: Mature technology. GPS and VMS are used as external observations to correct the output of the INS system.

Disadvantages: high costs. Depends on GPS, INS accuracy.

Scenario 3: The GPS matches the car map.

Advantages: Applied in car navigation.

Disadvantages: Depends on map accuracy and detail. Car navigation and accuracy.

Option 4: Correct the positioning and orientation sensor output, SLAM, by matching the environmental data.

Advantages: high precision, does not rely on high-precision positioning orientation sensor.

Disadvantages: Reliance on software algorithms, robustness and effectiveness need to be improved.

This article chooses option four.

(3) Body information module: including odometer and inertial sensor.

Over all,this article will choose follow sensors:

Positioning sensors including GPS locating the position of Latitude, Longitude and height,Inertial sensors (e.g. IMU) getting Angles (Roll, Pitch, Yaw) and Odometry collecting speed.

Environmental sensors including Camera which is stereo, texture (not used in race) ,Laser scanner (Lidar) which ranging with high resolution and Radar ranging with low resolution but distance, strong with dust and rain

There are four key problems to solve:

1.Multi-sensor information fusion - calibration

Obviously,comprehensive use of multiple sensors can improve sensing performance.However, whether or not the performance of the sensor can be improved depends on the sensor calibration and data fusion accuracy. High-performance multi-level data fusion processing requires high-precision coordinate system calibration.Meanwhile,the coordinate system calibration technology varies with sensors and data. The problem is the algorithm research of coordinate system calibration of complex sensor system is lack of literature.

2.Analysis of working environment of onboard sensors

There are bumpy jolts in the car's driving environment. Sometimes it encounters rain, snow, fog, and dust.

3. Design of sensor installation location

Understanding the principle of various sensors to ensure that the installation position and angle are in line with the calibration and the normal operation and service life of each sensor. Then use CATIA to design the sensor bracket.

4. Design of sensor sensing range

The sensor's comprehensive sensing range is 360 degrees without dead zone.

(1) Analyze the types of vehicle-based perception systems at home and abroad;

(2) Determine the vehicle perception system design scheme;

(3) Determine the technical parameters of the vehicle sensing system;

(4) Design and calculation of on-board sensing system;

(5) Determine the dimensions and parameters of the vehicle-mounted sensing system installation structure;

(6) Draw the general assembly drawing and main parts drawing of the vehicle sensing system.

参考文献

[1] 田甄.智能车辆多激光雷达目标检测系统的设计与实现[D].重庆邮电大学.2016

[2] 王宝锋，齐志权，马国成，陈思忠.一种基于雷达和机器视觉信息融合的车辆识别方法[J].汽车工程，Vol.37 ,No.6.2015

[3] 逄伟.低速环境下的智能车无人驾驶技术研究[D].浙江大学.2015

[4] 辛煜，梁华为，梅涛，黄如林等.基于激光传感器的无人驾驶汽车动态障碍物检测及表示方法[J].机器人，Vol.36, No.6.2014

[5] 白辰甲.基于计算机视觉和深度学习的自动驾驶方法研究[D].哈尔滨工业大学.2017

[6] 岳亚.基于深度相机人脸与行人感知系统的设计与实现[D].浙江大学.2017

[7] 石庭敏，蔡云飞，闫明.基于双多线激光雷达的低矮道边检测[J].计算机与数字工程,Vol.45 No.12.2017

[8] 刘继周.面向无人驾驶的智能车系统平台研究与应用[D].浙江大学.2017

[9] 王俊.无人驾驶车辆环境感知系统关键技术研究[D].中国科学技术大学.2016

[10] 崔佳超.无人驾驶智能车在动态环境中的避障方法[D].西安工业大学.2015

[11] Mokhtar Bouain , Karim M. A. Ali , Denis Berdjag,et al. An Embedded Multi-Sensor Data Fusion Design for Vehicle Perception Tasks[J].Journal of Communications Vol. 13, No. 1, January 2018

[12] Seong-Woo Kim, Baoxing Qin, Zhuang Jie Chong,et al.Multivehicle Cooperative Driving Using Cooperative Perception: Design and Experimental Validation[J].IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS.Volume: 16, Issue: 2, April 2015.

[13] Haiyan Shao, Kejie Li, Zhenhai Zhang.Study on Long-Distance Obstacle Perception of the Line Structured Light Sensor*[J].IEEE Conference on Robotics and Biomimetics .Zhuhai, China, December 6-9, 2015

[14] D. Martín , F. García, B. Musleh,et al.IVVI 2.0: An intelligent vehicle based on computational perception[J].Expert Systems with Applications 41 (2014) 7927–7944.

[15] Dan Alan Preston,Joseph David Preston,ET AL.SYSTEM AND METHOD FOR THE OPERATION OF AN AUTOMOTIVE VEHICLE SYSTEM WITH MODELED SENSORS[P].US 20170060810A1.