This PDF file contains the front matter associated with IS&T/SPIE proceedings volume 7539, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and the Conference Committee listing.

The Intelligent Ground Vehicle Competition (IGVC) is one of four unmanned systems student competitions that were founded by the Association for Unmanned Vehicle Systems International (AUVSI). The IGVC is a multidisciplinary exercise in product realization that challenges college engineering student teams to integrate advanced control theory, machine vision, vehicular electronics and mobile platform fundamentals to design and build an unmanned ground vehicle. Teams from around the world focus on developing a suite of dual-use technologies to equip their system of the future with intelligent driving capabilities. Over the past 17 years, the competition has challenged undergraduate, graduate and Ph.D. students with real world applications in intelligent transportation systems, the military and manufacturing automation. To date, teams from over 70 universities and colleges have participated. This paper describes some of the applications of the technologies required by this competition and discusses the educational benefits. The primary goal of the IGVC is to advance engineering education in intelligent vehicles and related technologies. The employment and professional networking opportunities created for students and industrial sponsors through a series of technical events over the four-day competition are highlighted. Finally, an assessment of the competition based on participation is presented.

The purpose of this paper is to discuss the challenge of engineering robust intelligent robots. Robust intelligent robots may be considered as ones that not only work in one environment but rather in all types of situations and conditions. Our past work has described sensors for intelligent robots that permit adaptation to changes in the environment. We have also described the combination of these sensors with a "creative controller" that permits adaptive critic, neural network learning, and a dynamic database that permits task selection and criteria adjustment. However, the emphasis of this paper is on engineering solutions which are designed for robust operations and worst case situations such as day night cameras or rain and snow solutions. This ideal model may be compared to various approaches that have been implemented on "production vehicles and equipment" using Ethernet, CAN Bus and JAUS architectures and to modern, embedded, mobile computing architectures. Many prototype intelligent robots have been developed and demonstrated in terms of scientific feasibility but few have reached the stage of a robust engineering solution. Continual innovation and improvement are still required. The significance of this comparison is that it provides some insights that may be useful in designing future robots for various manufacturing, medical, and defense applications where robust and reliable performance is essential.

This paper validates a previously introduced scalable modular control architecture and shows how it can be used to implement research equipment. The validation is conducted by presenting different kinds of micromanipulation applications that use the architecture. Conditions of the micro-world are very different from those of the macro-world. Adhesive forces are significant compared to gravitational forces when micro-scale objects are manipulated. Manipulation is mainly conducted by automatic control relying on haptic feedback provided by force sensors. The validated architecture is a hierarchical layered hybrid architecture, including a reactive layer and a planner layer. The implementation of the architecture is modular, and the architecture has a lot in common with open architectures. Further, the architecture is extensible, scalable, portable and it enables reuse of modules. These are the qualities that we validate in this paper. To demonstrate the claimed features, we present different applications that require special control in micrometer, millimeter and centimeter scales. These applications include a device that measures cell adhesion, a device that examines properties of thin films, a device that measures adhesion of micro fibers and a device that examines properties of submerged gel produced by bacteria. Finally, we analyze how the architecture is used in these applications.

Object detection and tracking play an increasing role in modern surveillance systems. Vision research is still confronted with many challenges when it comes to robust tracking in realistic imaging scenarios. We describe a tracking framework which is aimed at the detection and tracking of objects in real-world situations (e.g. from surveillance cameras) and in real-time. Although the current system is used for pedestrian tracking only, it can easily be adapted to other detector types and object classes. The proposed tracker combines i) a simple background model to speed up all following computations, ii)¹ a fast object detector realized with a cascaded HOG detector, iii) motion estimation with a KLT Tracker iv) object verification based on texture/color analysis by means of DCT coefficients and , v) dynamic trajectory and object management. The tracker has been successfully applied in indoor and outdoor scenarios it a public transportation hub in the City of Graz, Austria.

Dynamic obstacles like vehicles and animals can be distinguished from humans using their radar micro-Doppler signature. This allows customizing the robotic path algorithm to avoid highly sensitive and unpredictable obstacles like humans and rapidly moving obstacles like vehicles. We demonstrate the extraction of stride rate and other information associated with gait for enhanced person recognition from radar data. We describe the radar sensors used for the measurements, the algorithms used for the detection, tracking, and classification of people and vehicles, as well as describe some of the features that can be extracted. These features can serve as rudimentary identifying information in a scene with multiple subjects. We measure human subjects in indoor and outdoor clutter backgrounds for identification and gather ground truth using video to validate the radar data.

Ski jumping has continuously raised major public interest since the early 70s of the last century, mainly in Europe and Japan. The sport undergoes high-level analysis and development, among others, based on biodynamic measurements during the take-off and flight phase of the jumper. We report on a vision-based solution for such measurements that provides a full 3D trajectory of unique points on the jumper's shape. During the jump synchronized stereo images are taken by a calibrated camera system in video rate. Using methods stemming from video surveillance, the jumper is detected and localized in the individual stereo images, and learning-based deformable shape analysis identifies the jumper's silhouette. The 3D reconstruction of the trajectory takes place on standard stereo forward intersection of distinct shape points, such as helmet top or heel. In the reported study, the measurements are being verified by an independent GPS measurement mounted on top of the Jumper's helmet, synchronized to the timing of camera exposures. Preliminary estimations report an accuracy of +/-20 cm in 30 Hz imaging frequency within 40m trajectory. The system is ready for fully-automatic on-line application on ski-jumping sites that allow stereo camera views with an approximate base-distance ratio of 1:3 within the entire area of investigation.

This paper presents an online object tracking method, in which co-training and particle filters algorithms cooperate and complement each other for robust and effective tracking. Under framework of particle filters, the semi-supervised cotraining algorithm is adopted to construct, on-line update, and mutually boost two complementary object classifiers, which consequently improves discriminant ability of particles and its adaptability to appearance variants caused by illumination changing, pose verying, camera shaking, and occlusion. Meanwhile, to make sampling procedure more efficient, knowledge from coarse confidence maps and spatial-temporal constraints are introduced by importance sampling. It improves not only the accuracy and efficiency of sampling procedure, but also provides more reliable training samples for co-training. Experimental results verify the effectiveness and robustness of our method.

After utilizing robots for more than 30 years for classic industrial automation applications, service robots form a constantly increasing market, although the big breakthrough is still awaited. Our approach to service robots was driven by the idea of supporting lab personnel in a biotechnology laboratory. After initial development in Germany, a mobile robot platform extended with an industrial manipulator and the necessary sensors for indoor localization and object manipulation, has been shipped to Bayer HealthCare in Berkeley, CA, USA, a global player in the sector of biopharmaceutical products, located in the San Francisco bay area. The determined goal of the mobile manipulator is to support the off-shift staff to carry out completely autonomous or guided, remote controlled lab walkthroughs, which we implement utilizing a recent development of our computer vision group: OpenTL - an integrated framework for model-based visual tracking.

Submarine oil and gas pipeline inspection is a highly time and cost consuming task. Using an autonomous underwater vehicle (AUV) for such applications represents a great saving potential. However, the AUV navigation system requires reliable localization and stable tracking of the pipeline position. We present a method for robust pipeline localization relative to the AUV in 3D based on stereo vision and echo sounder depth data. When the pipe is present in both camera images, a standard stereo vision approach is used for localization. Enhanced localization continuity is ensured using a second approach when the pipe is segmented out in only one of the images. This method is based on a combination of one camera with depth information from the echo sounder mounted on the AUV. In the algorithm, the plane spanned by the pipe in the camera image is intersected with the plane spanned by the sea floor, to give the pipe position in 3D relative to the AUV. Closed water recordings show that the proposed method localizes the pipe with an accuracy comparable to that of the stereo vision method. Furthermore, the introduction of a second pipe localization method increases the true positive pipe localization rate by a factor of four.

A possibility to reduce the investment costs is an improved functionality of the inspection station in order to awaken investors' interest to buy. A new concept for 2-D and 3-D metric and logical quality monitoring with a more accurate, flexible, economical and efficient inspection station has been developed and tested at IITB. The inspection station uses short- and wide-range sensors, an intelligent grip system and a so-called "sensor magazine" which make the inspection process more flexible. The sensor magazine consists of various sensor ports with various task-specific, interchangeable sensors. With the sensor magazine and the improved measuring methods, the testing periods and therefore the costs in factory can be substantially decreased.

Increased interest in the exploration of extra terrestrial planetary bodies calls for an increase in the number of spacecraft landing on remote planetary surfaces. Currently, imaging and radar based surveys are used to determine regions of interest and a safe landing zone. The purpose of this paper is to introduce LandingNav, a sensor system solution for autonomous landing on planetary bodies that enables landing on unknown terrain. LandingNav is based on a novel multiple field of view imaging system that leverages the integration of different state of the art technologies for feature detection, tracking, and 3D dense stereo map creation. In this paper we present the test flight results of the LandingNav system prototype. Sources of errors due to hardware limitations and processing algorithms were identified and will be discussed. This paper also shows that addressing the issues identified during the post-flight test data analysis will reduce the error down to 1-2%, thus providing for a high precision 3D range map sensor system.

In this paper we describe a real-time 3D environment model for obstacle detection and collision avoidance with a mobile service robot. It is fully integrated in the experimental platform DESIRE. Experiments show, that all components perform well and allow for reliable and robust operation of a mobile service robot with actuating capabilities in the presence of obstacles.

One of the big challenges in multi-target tracking is the track management and correct data association between measurements and tracks. Major reason for tracking errors are detection failures such as merged, split, incomplete or missed detections as well as clutter-based detections (phantom objects). Those effects combined with uncertainties in existence and number of objects in the scene as well as uncertainties in their observability and dynamic object state lead to gross tracking errors. In this contribution we present an algorithm for visual detection and tracking of multiple extended targets which is capable of coping with occlusions and split and merge effects. Unlike most of the state-of-the-art approaches we utilize information about the measurements' composition gained through tracking dedicated feature points in the image and in 3D space, which allows us to reconstruct the desired object characteristics from the data even in the case of detection errors due to above-mentioned reasons. The proposed Feature-Based Probabilistic Data Association approach resolves data association ambiguities in a soft threshold-free decision based not only on target state prediction but also on the existence and observability estimation modeled as two additional Markov chains. A novel measurement reconstruction scheme allows for a correct innovation in case of split, merged and incomplete measurements realizing thus a detection-by-tracking approach. This process is assisted by a grid based object representation which offers a lower abstraction level of targets extent and is used for detailed occlusion analysis.

One of the most important problems in Computer Vision is the computation of the 2D projective transformation (homography) that maps features of planar objects in different images and videos. This computation is required by many applications such as image mosaicking, image registration, and augmented reality. The real-time performance imposes constraints on the methods used. In this paper, we address the real-time detection and tracking of planar objects in a video sequence where the object of interest is given by a reference image template. Most existing approaches for homography estimation are based on two steps: feature extraction (first step) followed by a combinatorial optimization method (second step) to match features between the reference template and the scene frame. This paper has two main contributions. First, for the detection part, we propose a feature point classification which is applied prior to performing the matching step in the process of homography calculation. Second, for the tracking part, we propose a fast method for the computation of the homography that is based on the transferred object features and their associated local rawbrightness. The advantage of this proposed scheme is a fast and accurate estimation of the homography.

This paper proposes a contribution for obstacles detection and tracking in the context of car-driving assistance. On the basis of the uv-disparity9 designed for in-vehicle stereovision, we propose a robust procedure to detect the objects located on the road plane. This detection is used as an initialization stage for a real-time template matching procedure based on the minimizing of a weighted cost function. An appropriate update of the weights, based upon the quality of the previous matching and depth information, allows to track efficiently non-rigid objects in front of a clutter background . Sequences involving pedestrians are used to demonstrate the efficiency of our procedure.

The performance of perceptive systems depends on a large number of factors. The practical problem during development is, that this dependency is very often not explicitly known. In this contribution we address this problem and present an approach to evaluate perception performance, as a function of e.g. quality of the sensor data. The approach is to use standardized quality metrics for imaging sensors, and to relate them to the observed performance of the environment perception. During our experiments, several imaging setups were analyzed. The output of each setup is processed offline to track down performance differences with respect to the quality of sensor data. We show how and to what extend the measurement of the Modulation Transfer Function (MTF) using standardized tests can be applied to evaluate the performance of imaging systems. The influence of the MTF on the signal-to-noise ratio can be used to evaluate the performance on a recognition task. We assess the measured performance by processing the data of different, simultaneously recorded imaging setups for the task of lane recognition.

Most current lens distortion models use only a few terms of Brown's model, which assumes that the radial distortion is dependant only on the distance from the distortion centre, and an additive tangential distortion can be used to correct lens de-centering. This paper shows that the characterization of lens distortion can be improved by over 79% compared to prevailing methods. This is achieved by using modern numerical optimization techniques such as the Leapfrog algorithm, and sensitivity-normalized parameter scaling to reliably and repeatably determine more terms for Brown's model. An additional novel feature introduced in this paper is to allow the distortion to vary not only with polar distance but with the angle too. Two models for radially asymmetrical distortion (i.e. distortion that is dependant on both polar angle and distance) are discussed, implemented and contrasted to results obtained when no asymmetry is modelled. A sample of 32 cameras exhibiting extreme barrel distortion (due to their 6.0mm focal lengths) is used to show that these new techniques can straighten lines to within 7 hundredths of a pixel RMS over the entire image.

This paper addresses the design, design method, test platform, and test results of an algorithm used in autonomous navigation for intelligent vehicles. The Bluefield State College (BSC) team created this algorithm for its 2009 Intelligent Ground Vehicle Competition (IGVC) robot called Anassa V. The BSC robotics team is comprised of undergraduate computer science, engineering technology, marketing students, and one robotics faculty advisor. The team has participated in IGVC since the year 2000. A major part of the design process that the BSC team uses each year for IGVC is a fully documented "Post-IGVC Analysis." Over the nine years since 2000, the lessons the students learned from these analyses have resulted in an ever-improving, highly successful autonomous algorithm. The algorithm employed in Anassa V is a culmination of past successes and new ideas, resulting in Anassa V earning several excellent IGVC 2009 performance awards, including third place overall. The paper will discuss all aspects of the design of this autonomous robotic system, beginning with the design process and ending with test results for both simulation and real environments.

A single-camera stereo vision system offers a greatly simplified approach to image capture and analysis. In the original study by Lovegrove and Brame, they proposed a novel optical system that uses a single camera to capture two short, wide stereo images that can then be analyzed for real-time obstacle detection. In this paper, further analysis and refinement of the optical design results in two virtual cameras with perfectly parallel central axes. A new prototype camera design provides experimental verification of the analysis and also provides insight into practical construction. The experimental device showed that the virtual cameras' axes possessed a deviation from parallel of less than 12 minutes (or 0.2°). The calculation of distances to objects from the two overlapping images all showed errors smaller than the pixel resolution limitation. In addition, a barrel lens correction was used in processing the image to allow parallax distance determination in the whole horizontal view of the images.

In this paper we address the problem of autonomous robot navigation in a "roadway" type environment, where the robot has to drive forward on a defined path that could be impeded by the presence of obstacles. The specific context is the Autonomous Challenge of the Intelligent Ground Vehicle Competition (www.igvc.org). The task of the path planner is to ensure that the robot follows the path without turning back, as can happen in switchbacks, and/or leaving the course, as can happen in dashed or single lane line situations. A multi-behavior path planning algorithm is proposed. The first behavior determines a goal using a center of gravity (CoG) computation from the results of image processing techniques designed to extract lane lines. The second behavior is based on developing a sense of the current "general direction" of the contours of the course. This is gauged based on the immediate path history of the robot. An adaptive-weight-based fusion of the two behaviors is used to generate the best overall direction. This multi-behavior path planning strategy has been evaluated successfully in a Player/Stage simulation environment and subsequently implemented in the 2009 IGVC. The details of our experience will be presented at the conference.

In this paper, we present Argos, an autonomous ground robot built for the 2009 Intelligent Ground Vehicle Competition (IGVC). Discussed are the significant improvements over its predecessor from the 2008 IGVC, Kratos. We continue to use stereo vision techniques to generate a cost map of the environment around the robot. Lane detection is improved through the use of color filters that are robust to changing lighting conditions. The addition of a single-axis gyroscope to the sensor suite allows accurate measurement of the robot's yaw rate and compensates for wheel slip, vastly improving state estimation. The combination of the D* Lite algorithm, which avoids unnecessary re-planning, and the Field D* algorithm, which allows us to plan much smoother paths, results in an algorithm that produces higher quality paths in the same amount of time as methods utilizing A*. The successful implementation of a crosstrack error navigation law allows the robot to follow planned paths without cutting corners, reducing the chance of collision with obstacles. A redesigned chassis with a smaller footprint and a bi-level design, combined with a more powerful drivetrain, makes Argos much more agile and maneuverable compared to its predecessor. At the 2009 IGVC, Argos placed first in the Navigation Challenge.

This paper presents the application of a distributed systems architecture to an autonomous ground vehicle, Q, that participates in both the autonomous and navigation challenges of the Intelligent Ground Vehicle Competition. In the autonomous challenge the vehicle is required to follow a course, while avoiding obstacles and staying within the course boundaries, which are marked by white lines. For the navigation challenge, the vehicle is required to reach a set of target destinations, known as way points, with given GPS coordinates and avoid obstacles that it encounters in the process. Previously the vehicle utilized a single laptop to execute all processing activities including image processing, sensor interfacing and data processing, path planning and navigation algorithms and motor control. National Instruments' (NI) LabVIEW served as the programming language for software implementation. As an upgrade to last year's design, a NI compact Reconfigurable Input/Output system (cRIO) was incorporated to the system architecture. The cRIO is NI's solution for rapid prototyping that is equipped with a real time processor, an FPGA and modular input/output. Under the current system, the real time processor handles the path planning and navigation algorithms, the FPGA gathers and processes sensor data. This setup leaves the laptop to focus on running the image processing algorithm. Image processing as previously presented by Nepal et. al. is a multi-step line extraction algorithm and constitutes the largest processor load. This distributed approach results in a faster image processing algorithm which was previously Q's bottleneck. Additionally, the path planning and navigation algorithms are executed more reliably on the real time processor due to the deterministic nature of operation. The implementation of this architecture required exploration of various inter-system communication techniques. Data transfer between the laptop and the real time processor using UDP packets was established as the most reliable protocol after testing various options. Improvement can be made to the system by migrating more algorithms to the hardware based FPGA to further speed up the operations of the vehicle.

An enhanced dynamic Delaunay Triangulation-based (DT) path planning approach is proposed for mobile robots to plan and navigate a path successfully in the context of the Autonomous Challenge of the Intelligent Ground Vehicle Competition (www.igvc.org). The Autonomous Challenge course requires the application of vision techniques since it involves path-based navigation in the presence of a tightly clustered obstacle field. Course artifacts such as switchbacks, ramps, dashed lane lines, trap etc. are present which could turn the robot around or cause it to exit the lane. The main contribution of this work is a navigation scheme based on dynamic Delaunay Triangulation (DDT) that is heuristically enhanced on the basis of a sense of general lane direction. The latter is computed through a "GPS (Global Positioning System) tail" vector obtained from the immediate path history of the robot. Using processed data from a LADAR, camera, compass and GPS unit, a composite local map containing both obstacles and lane line segments is built up and Delaunay Triangulation is continuously run to plan a path. This path is heuristically corrected, when necessary, by taking into account the "GPS tail" . With the enhancement of the Delaunay Triangulation by using the "GPS tail", goal selection is successfully achieved in a majority of situations. The robot appears to follow a very stable path while navigating through switchbacks and dashed lane line situations. The proposed enhanced path planning and GPS tail technique has been successfully demonstrated in a Player/Stage simulation environment. In addition, tests on an actual course are very promising and reveal the potential for stable forward navigation.

Using wide-angle or omnidirectional camera lenses to increase a mobile robot's field of view introduces nonlinearity in the image due to the 'fish-eye' effect. This complicates distance perception, and increases image processing overhead. Using multiple cameras avoids the fish-eye complications, but involves using more electrical and processing power to interface them to a computer. Being able to control the orientation of a single camera, both of these disadvantages are minimized while still allowing the robot to preview a wider area. In addition, controlling the orientation allows the robot to optimize its environment perception by only looking where the most useful information can be discovered. In this paper, a technique is presented that creates a two dimensional map of objects of interest surrounding a mobile robot equipped with a panning camera on a telescoping shaft. Before attempting to negotiate a difficult path planning situation, the robot takes snapshots at different camera heights and pan angles and then produces a single map of the surrounding area. Distance perception is performed by making calibration measurements of the camera and applying coordinate transformations to project the camera's findings into the vehicle's coordinate frame. To test the system, obstacles and lines were placed to form a chicane. Several snapshots were taken with different camera orientations, and the information from each were stitched together to yield a very useful map of the surrounding area for the robot to use to plan a path through the chicane.

A predictive object detection algorithm was developed to investigate the practicality of using advanced filtering on stereo vision object detection algorithms such as the X-H Map. Obstacle detection with stereo vision is inherently noisy and non linear. This paper describes the X-H Map algorithm and details a method of improving the accuracy with the Unscented Kalman Filter (UKF). The significance of this work is that it details a method of stereo vision object detection and concludes that the UKF is a relevant method of filtering that improves the robustness of obstacle detection given noisy inputs. This method of integrating the UKF for use in stereo vision is suitable for any standard stereo vision algorithm that is based on pixel matching (stereo correspondence) from disparity maps.

A mobile robot moving in an environment in which there are other moving objects and active agents, some of which may represent threats and some of which may represent collaborators, needs to be able to reason about the potential future behaviors of those objects and agents. In previous work, we presented an approach to tracking targets with complex behavior, leveraging a 3D simulation engine to generate predicted imagery and comparing that against real imagery. We introduced an approach to compare real and simulated imagery using an affine image transformation that maps the real scene to the synthetic scene in a robust fashion. In this paper, we present an approach to continually synchronize the real and synthetic video by mapping the affine transformation yielded by the real/synthetic image comparison to a new pose for the synthetic camera. We show a series of results for pairs of real and synthetic scenes containing objects including similar and different scenes.

The desirability and challenge of developing a completely autonomous vehicle and the rising need for more efficient use of energy by automobiles motivate this research- a study for an optimum solution to computer control of energy efficient vehicles. The purpose of this paper is to compare three control methods - mechanical, hydraulic and electric that have been used to convert an experimental all terrain vehicle to drive by wire which would eventually act as a test bed for conducting research on various technologies for autonomous operation. Computer control of basic operations in a vehicle namely steering, braking and speed control have been implemented and will be described in this paper. The output from a 3 axis motion controller is used for this purpose. The motion controller is interfaced with a software program using WSDK (Windows Servo Design Kit) as an intermediate tuning layer for tuning and parameter settings in autonomous operation. The software program is developed in C++. The voltage signal sent to the motion controller can be varied through the control program for desired results in controlling the steering motor, activating the hydraulic brakes and varying the vehicle's speed. The vehicle has been tested for its basic functionality which includes testing of street legal operations and also a 1000 mile test while running in a hybrid mode. The vehicle has also been tested for control when it is interfaced with devices such as a keyboard, joystick and sensors under full autonomous operation. The vehicle is currently being tested in various safety studies and is being used as a test bed for experiments in control courses and research studies. The significance of this research is in providing a greater understanding of conventional driving controls and the possibility of improving automobile safety by removing human error in control of a motor vehicle.

When a standard ND (N-dimension) curve is compared to a test ND curve, the Euclidean distance between these two curves is defined as the root-mean-square sum of the distances between each (test) point in the test curve and the point on the standard curve closest to that test point. This number of root-mean-square sum is called the closeness CLS of the test curve to the standard curve in the ND state space where each point in the space may represent a snap shot of a certain moving or time-varying object. This CLS number can be used to identify accurately, a test moving object against several standard moving objects stored in the memory of this dynamic pattern recognition system, according to the particular way each standard object moves. In this paper, we use the face recognition scheme as a particular example that leads to the general analysis and design procedure.

Many agricultural non-contact visual inspection applications use color image processing techniques because color is often a good indicator of product quality. Color evaluation is an essential step in the processing and inventory control of fruits and vegetables that directly affects profitability. Most color spaces such as RGB and HSV represent colors with three-dimensional data, which makes using color image processing a challenging task. Since most agricultural applications only require analysis on a predefined set or range of colors, mapping these relevant colors to a small number of indexes allows simple and efficient color image processing for quality evaluation. This paper presents a simple but efficient color mapping and image processing technique that is designed specifically for real-time quality evaluation of Medjool dates. In contrast with more complex color image processing techniques, the proposed color mapping method makes it easy for a human operator to specify and adjust color-preference settings for different color groups representing distinct quality levels. Using this color mapping technique, the color image is first converted to a color map that has one color index represents a color value for each pixel. Fruit maturity level is evaluated based on these color indices. A skin lamination threshold is then determined based on the fruit surface characteristics. This adaptive threshold is used to detect delaminated fruit skin and hence determine the fruit quality. The performance of this robust color grading technique has been used for real-time Medjool date grading.

Spline representations have been successfully used with a genetic algorithm to determine a disparity map for stereo image pairs. This paper describes work to modify the genetic spline algorithm to use a version of the genetic algorithm with small populations and few generations, previously referred to as "Tiny GAs", to allow algorithm implementations to achieve real-time performance. The algorithm was also targeted at unrectified stereo image pairs to reduce preprocessing making it more suitable for real-time performance. To ensure disparity map quality is preserved, the two dimensional nature of images is maintained to leverage persistent information instead of representing the images as 1-D signals as suggested in the orignal genetic spline algorithm. Experimental results are given of this modified algorithm using unrectified images.

A feature tracker is only as good as the features found by the feature detector. Common feature detectors such as Harris, Sobel, Canny, and Difference of Gaussians convolve an image with a specific kernel in order to identify "corners" or "edges". This convolution requires, however, that the source image contain only one value (or color channel) per pixel. This requirement has reduced the scope of feature detectors, trackers, and descriptors to the set of gray scale (and other single-channel) images. Due to the standard 3-channel RGB representation for color images, highly useful color information is typically discarded or averaged to create a gray scale image that current detectors can operate on. This removes a large amount of useful information from the image. We present in this paper the color Difference of Gaussians algorithm which outperforms the gray scale DoG in number and quality of features found. The color DoG utilizes the YCbCr color space to allow for separated processing of intesity and chrominance values. An embedded vision sensor based on a low power field programmable gate array (FPGA) platform is being developed to process color images using the color DoG with no reduction in processing speed. This low power vision sensor will be well suited for autonomous vehicle applications where size and power consumption are paramount.

To improve the identification rate and tracking rate for quickly moving target, expand tracking scope and lower the sensitivity to illumination varying, an active visual tracking system self-adapting to illumination based on particle filter pre-location is proposed. The algorithm of object pre-location based on particle filter is used to realize realtime tracking to moving target by forecasting its location and control camera joints of Tilt and Pan. The method resetting system is used to improve accuracy of system. Brightness histogram equalization method is used to reduce the affect of illuminating varying in pre-location algorithm. Experiments and property analysis show that the real-time and accuracy are greatly improved.