Convolutional Neural Network (CNN) Approach for Enhancing the Identification of UWB Radar Targets

Get Complete Project Material File(s) Now! »

Key Technologies in Autonomous Vehicle

This technology originated from multi-disciplinary scientific researches such as vehicle control systems, robotics, sensors, machine perception, innovation, mapping and machine learning [30]. The important keys in creation of this autonomous vehicle are telecommunication and navigation systems, sensors technologies and artificial intelligence systems. These core technologies can be derived into four parts based on the functionality of self driving car [31]: environment perception, car navigation, path planning, and the car control.

Environmental Perception Systems

Perceiving surrounding environment, so that the drivers can interact consequently, is the key success of reducing the road traffic accident. As earlier mentioned in the chapter 1, the focus of this thesis work is on developing a system that the bus can perceive the movement of the VRUs particularly in the blind spot areas, so that the system can give spatial alert to the bus driver to be aware of the presence of the VRUs around them. Therefore, related to this topic, there are two main concerns of the discussion in this section: the anti-collision and the Vulnerable Road Users (VRUs) protection systems and applications.

Systems and Applications of Anti-Collision

In recent years, the obstacle and collision avoidance have attracted a lot of researcher’s attention. With the aim of developing a technology that can facilitate to prevent or to reduce road accidents due to human error and to make drivers comfort, this technology has become very popular in these days. In order to achieve a good level precision, there are two difference technologies approach have been used, passive and active sensor system [32] [33] that it depends on the applications. In passive technology, a sensor receives signals coming from environment. This kind of sensors includes in particular stereo-cameras or optical-flow cameras [33]. While, the active system works based on the response (information gathered) of transmitted signals from a sensor to analyze the actual situation.
There are many research projects and applications have been conducted and developed relating to Anti-collision systems and the some of them will be discussed in the following paragraphs. In [34], they used a combination of the laser beams and camera to estimate position of the vehicle after 1 second, then project it on the road surface. The installed camera on the vehicle is used to capture the positions of each laser beams. The obstacles can be detected by measuring the difference between normal and abnormal road conditions. In [35], an Anti-Collision System (ACS) that allows Automatic breaking system, has been developed by using a stereo multi-purpose camera and ultrasonic sensor. The camera has been used to get dimensional data of the vehicle and its environment and the obstacle is detected by ultrasonic sensor. These information are then sent to an automated emergency braking system that breaks the vehicle at an appropriate time and condition. In [36], they developed a collision avoidance system in heavy traffic where the vehicles speed normally less than about 20 km/h. This system puts 8 ultra-sonic sensors to cover all around the car. In [37], they reported that an autonomous cruise control (ACC) system was firstly in-troduced by Mercedes-Benz in 1999 for their S-class series car. This ACC is a 77 GHz radar-based. Since then, several type of radar system have emerged in developing the au-tonomous car system. They are based mainly on one of the following technologies: FMCW, FSK-CW (Frequency Shift-Keying-Continuous Wave) and MFSK-CW (Multiple Frequency Shift Keying-Continuous Wave). For the Short Range Radar (SRR) technologies mainly use UWB technologies: UWB pulse, spread spectrum techniques and radar « Stepped Frequency ».
National Highway Traffic Safety Administration U.S. Department of Transportation en-tered a collaborative research project in Advanced Collision Avoidance System/Field Oper-ational Test (ACAS/FOT) Program. The members are: Delphi-Delco Electronic Systems, Delphi Chassis, HRL Laboratories, HE Microwave and UMTRI. In ACAS the FCW (Forward Collision Warning) and ACC (Adaptive Cruise Control) functions are implemented using a combination of a long-range forward radar-based sensor, a forward vision-based sensor and a Global Positioning Satellite (GPS). The objective of this research is to determine the practi-cal suitability of the combined ACC/FCW function for widespread use by the driving public [38].

Systems and Applications of VRUs Protection

Many techniques have been proposed over the years in order to reduce the injuries and mor-tality caused by road crash accidents including the video and radar techniques. PROTEC-TOR, SAVE-U (Sensors and system architecture for Vulnerable road Users Protection) and PROSPECT (Proactive Safety for Pedestrians and Cyclists) were the EU projects concerned in protecting the uncovered road users. SAVE-U concerned in developing sensor-based driver assistance system by integrating three different technologies of sensor (radar, IR, camera) simultaneously using sensor fusion to optimize VRUs (vulnerable Road Users) detection [39]. In the PROTECTOR, they focused also on three sensors. Beside the camera and laser scan-ner, the microwave radar was used for obstacle detection, around the vehicle. Using this kind of radar, detection of pedestrian among the other objects is done by evaluating the reflected power, the power variant over tunnel, the dimension and prevailing dynamic of the obstacle [40]. PROSPECT is a collaborative research project funded by the European Commission, aimed at improving the protection of vulnerable road users (VRUs) with an emphasis on the two groups with the largest shares of fatalities: cyclists and pedestrians. The project started in May 2015 and involves many relevant partners from the automotive industry, academia and independent test labs [41].
The PROTECTOR and SAVE-U exploited the microwave 24 GHz narrow band radar and PROSPECT used 77 GHz narrow band high resolution radar. In one hand, using a narrow band radar requires a very complex process to extract information in order to classify correctly the type of radar target. On the other hand, using UWB radar is interesting to be investigated for this application because it has a rich transitory response thanks to its large bandwidth used. So, theoretically, it has a good capacity in target recognition. After reviewing some of the anti-collision and VRUs protection systems above and com-pared to another system like cameras, lidar and ultra-sonic, the radar system has some additional important advantages of robustness in all weather conditions, good estimation of target distance, and faster target discrimination. The radar systems are suited for every weather condition like foggy, rainy and etc. Therefore, our next discussions will be focused on the radar systems.

READ  Anisotropic Magnetoresistance Sensor 

Radar Systems

The Radar, RAdio Detection And Ranging, works by transmitting and receiving the elec-tromagnetic waves. The analysis of their reflections is used to identify the presence and determine the parameters of objects located in the environment (position, speed, direction of movement, etc). Since the Second World War, radar technologies have been the subject of many industrial developments. While initially the applications were purely military, many consumer applications have emerged in aeronautics, shipping, robotics, meteorology or driv-ing. Radar transmits and receives electromagnetic waves which aims to determine the range, velocity and altitude of the target objects. Basically, the radar has two main different types, that are narrow band radars and wide-band radars. The narrow band radars are mostly used in long range applications and the wide-band radars are mostly used in short range applications. The VRUs protection systems normally need a short range radar with a high precision. In this case the UWB radar is suitable for this application.

Table of contents :

1.1 Background
1.2 Problematics
1.3 CYCLOPE Project
1.3.1 Brief Introduction
1.3.2 Relation between Thesis Work and CYCLOPE Project
1.4 The General Research Context
1.5 Thesis Scope
1.6 Methodology
1.7 Thesis Contributions
1.8 Thesis Organization
1.9 List of Publications
2.1 Intelligent Transportation System (ITS)
2.1.1 Applications of ITS
2.1.2 Key Underlying ITS Technologies
2.2 Overview of Autonomous Vehicles
2.2.1 Autonomous Vehicles Operational Levels
2.2.2 Key Technologies in Autonomous Vehicle
2.3 Environmental Perception Systems
2.3.1 Systems and Applications of Anti-Collision
2.3.2 Systems and Applications of VRUs Protection
2.4 Radar Systems
2.4.1 Radar Equation
2.4.2 Range Resolution
2.4.3 Different type of Radar
2.5 Ultra-Wide Band Radar
2.5.1 Ultra-Wide Band Technology
2.5.2 UWB Waveform
2.5.3 Impulse UWB Radar
2.6 Radar Detector and Noise Removal Method
2.7 Conclusion of Chapter 2
3.1 UWB Radar Module
3.2 Detection Theory
3.2.1 Probability of Detection and Probability of False Alarm
3.2.2 Neyman-Pearson Criteria
3.3 Development of UWB Radar Detector
3.3.1 Moving Target Indication (MTI)
3.3.2 Higher Order Statistics
3.3.3 Time Delay Estimation
3.3.4 Proposed UWB Radar Detector
3.4 Conclusion of Chapter 3
4.1 Radar Signature
4.1.1 Obtaining Radar Signature
4.2 SVM-Based Approach
4.2.1 Solving optimization problem
4.2.2 SVM kernel
4.2.3 Experimental SVM Results
4.3 A Deep Belief Network Approach
4.3.1 Introduction of Artificial Neural Networks
4.3.2 Restricted Boltzmann Machine (RBM)
4.3.3 Training Deep Belief Network
4.3.4 Results and Discussion
4.3.5 Performance Comparison of DBN and SVM
4.4 Conclusion of Chapter 4
5.1 Study of Noise Removal Techniques
5.1.1 Principal Component Analysis (PCA)
5.1.2 Singular Value Decomposition (SVD)
5.1.3 Wavelet Shrinkage Denoising (WSD)
5.1.4 Combination of HOS and WSD (Proposed Method)
5.1.5 Results and Discussion
5.2 Convolutional Neural Network (CNN) Approach for Enhancing the Identification of UWB Radar Targets
5.2.1 Preprocessing 2-D Radar Data
5.2.2 Convolution Neural Network
5.2.3 Results and Discussions
5.2.4 Validation of the Developed System
5.3 Conclusion of Chapter 5


Related Posts