This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
This research project was initiated following consultation with Jaguar Land Rover and it looks at how the Terrain Response system on Land Rovers can be automated. The research is aimed at developing a system capable of identifying and distinguishing between different terrains. Land Rover's patented Terrain Response (TR) system comes as standard on all newer models and it allows the driver to adjust the chassis and transmission settings of the vehicle to suit the terrain they are travelling on. Since the research project is specific to Land Rover's patented Terrain Response system, extensive consultation with Jaguar Land Rover was essential in order to establish key issues of importance in the operation of the Terrain Response system.
Following extensive literature review and consultation with experts at Jaguar Land Rover it was established that the driving experience would be further improved by automating the Terrain Response system. From studying previous related research work, it was also establish that different terrains can be distinguished by capturing live images of the terrain the vehicle is travelling on, using a forward looking camera and then using computer vision processes to apply image segmentation based on colour.
The research project illustrates a real time application of computer vision and image processing methods. It aims to automate the TR system which currently requires input from the driver, thus at the same time improving the driving experience.
Figures and Tables v
1.1 Background 1
1.2 Terrain Detection 3
1.3 Aims and Objectives 6
2 LITERATURE REVIEW
2.1 Technology Review 9
2.2 Human Vision System 10
2.3 Influencing Factors on Methodology 11
2.3.1 Technology 11
2.3.2 Environment 13
2.3.3 Hardware 15
3 RESEARCH METHODOLOGY
3.1 Detailed Analysis 16
3.2 System Operation 18
3.3 Outcome and Discussion 20
4.1 Conclusions 21
4.2 Suggestions for Future Work 24
Appendix 1: Screenshot of the Software Front-End 29
Appendix 2: Tables of Technical Constants 32
The main source of inspiration behind this project is the need to understand and also demonstrate the various and diverse applications of Computer Vision. Computer vision is the science and technology of machines or artificial systems that are able to extract information from image data. The image data can be captured in different forms which include, video sequences, views from multiple cameras, or even multi-dimensional data from a medical scanner. Computer vision systems can be derived from theories and models of computer vision technology. There are many existing examples of applications of computer vision systems such as, controlling processes, detecting events, organising information, modelling objects or environments, and computer-human interaction. There are many application fields for computer vision with the medical field being the most prominent. In this case study the main focus will be on object recognition. There are four main processes of computer vision which will be of paramount importance in this exercise. These are image acquisition, image distribution, image processing, and image analysis .
Image segmentation (processing) is an important process in many areas of computer vision like object recognition and stereo vision. Image segmentation and object recognition will form the core areas of research in this case study. Image segmentation provides further information about the properties of an image by identifying regions of similar colour, texture and also the edges of an image. This process enables simplification of an image from thousands of pixels to relatively few segments which can be identified and distinguished. Specific portions of an image can be removed or isolated by applying image segmentation .
1.2 Terrain Detection
Terrain Detection (TD) involves identifying the outlay of the surface the vehicle is travelling on. There are several ways of identifying or detecting the nature of a terrain which include visual and ultra sound.
In order to achieve automatic control of the TR system which now comes as standard on all Land Rover four wheel drive vehicles, terrain detection becomes essential. The TR system has five settings which are available via a rotary knob on the centre
Fig 1.1: Land Rover's TR system control knob4
console, which include general driving; grass, gravel and snow; mud and ruts; sand; and rock crawl. The TR system will control suspension ride height, engine management, throttle mapping, transfer case ranges, and transmission settings when active. Electronic Driving Aids (EDA) such as electronic traction control (ETC), dynamic stability control (DSC) and hill descent control (HDC), and electronic e-diffs are also manipulated through the TR system . With its breadth of capabilities, Terrain Response is designed to enhance the drivers' everyday driving experience whilst at the same time successfully tackling extreme off-road conditions with ease.
Mobile robots have been used for decades to explore planetary surfaces on Earth and in space, as a means of reducing cost or the hazard of human intervention. When a situation arises whereby there is limited direct human control over robots, it is critical to improve the robotic system performance. The ability to distinguish distinct terrain classes is important to enable movement of the Robot. Mobile robots are usually programmed to recognise some known terrain classes, and also distinguish them from unknown terrain. When it comes to scientific research, terrain classification in any terrestrial environment is necessary in order for a mobile robot to achieve its objectives . Autonomous navigation in natural terrain is an emerging technology area of interest for purposes of military surveillance, reconnaissance, and target tracking applications. In particular NASA is interested with this technology sector for its robotic planetary space exploration missions.
In the future, robotic military and space applications will require more intelligent mobile robots with advanced terrain adaptability. The endurance of robots during extended terrain navigation missions, in natural terrain environments is dependent upon the robots' ability to perceive its environment consistently and also adjust to it dynamically. The control scheme of the robot would endure a lot of challenges whilst attempting to navigate the robot safely without terrain adaptability. Terrain adaptability requires carefully considered terrain assessment and situation-awareness .
1.3 Aims and Objectives
To demonstrate an application of image processing.
To demonstrate understanding of computer vision processes.
To further improve driving experience of Land Rover's 4x4's.
To improve safety during driving.
To facilitate further development of Land Rover's patented TR system.
By applying computer vision techniques like image segmentation in order identify features of interest which relate to terrain classifications.
By adding a degree of automation to the knob control action of the TR system, thus eliminating the need for the driver to manually adjust the TR system control.
2 LITERATURE REVIEW
2.1 Technology Review
The technology which is required for the project to achieve the desired results has been around for years. It is essential to demonstrate a good understanding of the theory as well as showing its applicability. Image segmentation has been around for a while now and understanding its application enables simplification of many tasks and also increases productivity. The aim of using image segmentation is to change the representation of an image into something that is understandable and relatively easier to analyse . There are several ways of applying image segmentation and they produce differing results. The outcome from image segmentation is a set of segments that collectively fill the entire image. The result of segmentation can also be a set of contours or boundaries extracted from the image, and each one of the pixels in a particular region are similar with respect to some characteristic or computed property. They can possess similar properties like colour, intensity or texture. This thereby means that adjacent regions or segments will be noticeably different with respect to their characteristic or computed properties.
When it comes to TD the technology involved must be able extrapolate relevant information or characteristics from the area of interest. Image acquisition is one of the most important processes when it comes to TD. Image generating devices range from charge-coupled devices (CCD) to ultrasonic imagers. Special image sensors are used for various applications such as thermal imaging, creation of multi-spectral images, gamma cameras, sensor arrays for x-rays, and other highly sensitive arrays for astronomy. CCD cameras and sensors generally make up the largest portion of digital cameras on the market. CCDs are image sensors that consist of a combination of shift registers, which form a two-dimensional array. These shift register cells contain a charge that relates to the brightness at the respective CCD cell, which will then be represented by a picture element (pixel). The concept of shift registers makes it relatively easy for the parallel-serial conversion to take place by shifting the values to the sensor bottom and then reading one information line , as shown in Fig 2.1.
Line-scan cameras can also be used as an image generating sensor to acquire two-dimensional images. A line-scan camera contains an image sensor chip and the focusing mechanism. The image sensor moves along the object being scanned, similar to the principle of operation of a flat-bed image scanner, whereby 2D images are transferred from paper into a digital media format. With line-scan camera technology it is relatively easy to generate images with a large number of pixels. Line-scan cameras are mostly used in industrial settings during product inspection mainly to capture an image of a constant stream of moving material on a conveyor belt or rotating cylindrical objects.
Another image sensor technology that is available on the market is structured like a CMOS (Complementary Metal-Oxide-Semiconductor) memory chip, therefore it needs row and column address decoders. CMOS is a technology used for constructing integrated circuits, and it is integrated into microprocessors, microcontrollers, static RAM, digital logic circuits, and also analog circuits such as image sensors. CMOS image sensors make use of a photon-to-electron-to-voltage conversion, which takes place in each picture element. This makes it possible to analyse only a region of interest by manipulating the row and column address decoders of the region. By integrating further components on the image sensor chip, like clocking and timing, this makes the camera more reliable and cheaper.
2.2 Human Vision System
To illustrate how far computer vision systems have to go before matching human ability, it is important to offer a brief description of the human visual system. When it comes to image acquisition there is no image capturing device yet developed that comes even close to the human eye's superiority . Basically eyes are organs that detect light and then send electrical impulses along the optic nerve system to the visual and other parts of the brain.
Fig 2.2: Schematic diagram of the human eye 1
Images are projected onto the retina of the eye, and the retina contains two major types of light-sensitive photo-receptor cells used for vision (cones and rods). Rods are unable to distinguish colours, but they are responsible for low-light, monochrome vision. Rods operate effectively in dim light since they contain a pigment, visual purple (rhodopsin), which is sensitive at low light intensity, but it saturates at higher intensities. The density of Rods is greater in the peripheral retina than it is in the centre of the retina. So the cones are responsible for colour vision . Most cones are located in the central part of the retina, known as the fovea. Each cone is connected by a nerve to the brain, and they are sensitive to bright light hence the cones do not work in dim light.
The eye can adapt to light intensities of the order 1010, ranging from the lowest visible light to the highest bearable intensity. The eye can cope with a wide range of light levels by altering its own sensitivity depending on the level of brightness of the light. However the eye cannot distinguish between different dim and different bright ambient levels at the same time. Assuming that the brightness perceived by the eye is a logarithmic function of intensity, a small increase in dim light intensity can be equated to a large increase in bright light intensity.
Since the perception of brightness is a function of wavelength, it can be analysed using the relative luminosity function, as shown in Fig 2.3. From the relative luminosity function it can deduced that light radiating at mid-spectrum wavelength will appear brighter than light at either end spectrum wavelength .
2.3 Influencing Factors on Design Methodology
A lot of things influence choice of design methodology. The final design system will depend on the type of data to be extrapolated, technology accessibility and cost. In order to optimise the design choice, careful consideration must also be placed on the relative cost of the hardware and software.
The technology required to produce the design proposal is readily available on the market, and the choice of technology is crucial. When it comes to choice of technology the area most affected is the image acquisition process and image processing. In term of image acquisition the technologies which would most effectively produce the required result are CCD and CMOS cameras. Table 2.1 below shows a comparison of the main features of the technology.
Only full image accessible
Single pixel access possible
Relatively high power consumption
Low power consumption
Only image sensor on chip
Additional features on chip possible
Full image, after charge transfer
At each pixel
Fixed pattern noise and statistical noise
Additional noise sources to CCD
Anti-blooming techniques available
Natural blooming immunity
Smear caused by charge transfer
Table 2.1: Comparison of CCD and CMOS technology 1
The technology for image processing is now widely available via add on boards for microprocessors which perform fast convolutions and fast Fourier transformations. Many leading micro chip manufactures have developed advanced systems dedicated to image processing in real time.
Another challenging task in the design process is being able to come up with hardware and software which is adaptive to the different operating conditions. Since the changes in terrain are mainly influenced by different weather conditions, it is essential for the technology to adapt to the different conditions.
In order to maintain precise control in difficult driving conditions, the drive has to select one of the five intuitive programs which are uniquely tailored to different terrains: (1) Dynamic Program (General driving on earlier models); (2) Grass, Gravel and Snow; (3) Mud and Ruts; (4) Sand; or (5) Rock Crawl. Engineers at Land Rover studied 50 different driving surfaces and identified that they could be categorised into five unique terrain conditions whether on or off-road . From the categorised settings it is apparent that the nature of the environment or driving conditions will become an important issue in the design process.
3 RESEARCH METHODOLOGY
3.1 Image Acquisition
The initial step of any computer vision system is the image acquisition stage, after which various techniques of image processing can be applied. If the image has not been properly acquired then the intended image processing tasks may not be achievable, even with the aid of image enhancement.
3.1.1 Image Texture
The terrain can be classified based on the relevant features extracted from images of the driving surface. Terrain condition assessment can be achieved by performing image-based object surface texture analysis. Objects in different driving terrain have some distinguishing visual texture properties. There are certain surface texture attributes common in the images which include variance, energy, contrast, entropy, and homogeneity. Image texture is a result of geometrical organisation or it can be due to the occurrence in repeated designs of certain pixel intensity variations. The repeated designs are instigated by illumination of terrain object surfaces such as scattered rocks, grass, sand or snow. Differences in surface texture patterns can be used to regularly classify the condition of a terrain and determine the best TR option .