This thesis examines a complete design framework for a real-time, autonomous system with specialized VLSI hardware for computing 3-D camera motion. In the proposed architecture, the first step is to determine point correspondences between two images. Two processors, a CCD array edge detector and a mixed analog/digital binary block correlator, are proposed for this task. The report is divided into three parts. Part I covers the algorithmic analysis; part II describes the design and test of a 32$\time $32 CCD edge detector fabricated through MOSIS; and part III compares the design of the mixed analog/digital correlator to a fully digital implementation.

We analyze videos in which a hand interacts with a basketball. In this work, we present a computational system which detects and classifies hand-ball events, given the trajectories of a hand and ball. Our approach is to determine non-gravitational parts of the ball's motion using only the motion of the hand as a reliable cue for hand-ball events. This thesis makes three contributions. First, we show that hand motion can be segmented using piecewise fifth-order polynomials inspired by work in motor control. We make the remarkable experimental observation that hand-ball events have a phenomenal correspondence to the segmentation breakpoints. Second, by fitting a context-dependent gravitational model to the ball over an adaptive window, we can isolate places where the hand is causing non-gravitational motion of the ball. Finally, given a precise segmentation, we use the measured velocity steps (force impulses) on the ball to detect and classify various event types.

We analyze videos in which a hand interacts with a basketball. In this work, we present a computational system which detects and classifies hand-ball events, given the trajectories of a hand and ball. Our approach is to determine non-gravitational parts of the ball's motion using only the motion of the hand as a reliable cue for hand-ball events. This thesis makes three contributions. First, we show that hand motion can be segmented using piecewise fifth-order polynomials inspired by work in motor control. We make the remarkable experimental observation that hand-ball events have a phenomenal correspondence to the segmentation breakpoints. Second, by fitting a context-dependent gravitational model to the ball over an adaptive window, we can isolate places where the hand is causing non-gravitational motion of the ball. Finally, given a precise segmentation, we use the measured velocity steps (force impulses) on the ball to detect and classify various event types.

Municipalities and contractors in Canada and other parts of the world rely on road surface condition information during and after a snow storm to optimize maintenance operations and planning. With an ever increasing demand for safer and more sustainable road network there is an ever increasing demand for more reliable, accurate and up-to-date road surface condition information while working with the limited available resources. Such high dependence on road condition information is driving more and more attention towards analyzing the reliability of current technology as well as developing new and more innovative methods for monitoring road surface condition. This research provides an overview of the various road condition monitoring technologies in use today. A new machine vision based mobile road surface condition monitoring system is proposed which has the potential to produce high spatial and temporal coverage. The proposed approach uses multiple models calibrated according to local pavement color and environmental conditions potentially providing better accuracy compared to a single model for all conditions. Once fully developed, this system could potentially provide intermediate data between the more reliable xed monitoring stations, enabling the authorities with a wider coverage without a heavy extra cost. The up to date information could be used to better plan maintenance strategies and thus minimizing salt use and maintenance costs.

The development of machine vision algorithms for inspection and machine guidance has traditionally relied on the knowledge and experience of the developers as most of the techniques are based on heuristics and trial and error. This is especially problematic in the area of natural products where variability of the products is the rule rather than the exception. Humans are particularly good in functioning in this arena and in this thesis we look at the development of techniques derived from the functions of the human visual system (HVS). We first identify the significant processes in the HVS and highlight those that we believe are germane to the problems of interest. We then develop computational techniques using these methods and demonstrate their applicability to practical problems. This thesis uses the knowledge that the HVS is considered to consist of three sequential operations (sensing; encoding/transfer; and image interpretation) as a basis for developing a parallel procedure for a machine vision system. We have found that outputs derived from a simulation of the behaviors of receptive fields in the retina and in the higher levels of the brain can generate useful and robust features. Equivalent processes are then developed for machine applications under the guidance of a human operator to identify the areas of interest in the scene for the problem under consideration. Specifically we use the processes for encoding/transfer of data from the retina to develop techniques to enhance color contrasts, and compute color image features that are useful for defect detection and identification in real world images. This is accomplished by a transformation from image space to a characteristic response space that improves the robustness of classification. In this thesis the approach developed is applied to two industrial problems in the quality monitoring of meat and vegetables. The first problem concerns the quality monitoring of breast butterflies and the other the detection of defects on the surface of citrus. The approach is shown to derive algorithms that are robust and can be implemented at high rates of speed. Additionally we also identify a model within which further developments can be conducted as we learn more about the functioning of the HVS.

Abstract In the researches of image tracking to restore an object¡¦s position or velocity in the space, it is expectable that increasing numbers of camera can reduce the error. In fact, this phenomenon happens in practical applications. But so far, the physical theory behind this effect has not been fully known. Therefore, based on this motivation, this thesis tends to lay the physical foundation of specific machine vision problem using the statistical probability concept. Extensive error analysis and computer simulation for motion vector of translation movement solved by the least squares technique are conducted by incorporating Gaussian noised into optical flow components. It is expected to provide an effective theoretical model for further developments. Keywords¡GImage tracking, The least squares method, Gauss distribution, Error analysis.

This paper investigates two distinct, but related, topics in machine vision. The first is the effect of lighting on the performance of a 2D vision-based inspection system. The lighting component of machine vision has often been overlooked; an attempt was made to quantify the impact on existing machine vision algorithms. The second topic explores the applications of a data-rich 3D vision sensor that is capable of providing depth data in a wide range of ambient lightning conditions for industrial applications. A focus is placed on inspection systems with the depth data provided by the sensor. Three basic lighting geometries were compared quantitatively based on discriminant analysis in an inspection task that checked for the presence of J-clips on an aluminum carrier. Two different LabVIEW® machine vision algorithms were used to evaluate backlight, bright field and dark field illumination on their ability to minimize the span of the pass (clip present) and fail (clip absent) sample sets, as well as maximize the separation between these sample sets. Results showed that there were clear differences in performance with the different lighting geometries, with over a 30% change in performance. Although it has long been accepted that the choice of lighting for machine vision systems is not a trivial exercise, this paper provides a quantitative measure of the impact lighting has on the performance of feature-based machine vision. The Microsoft Kinect® is a commercial vision sensor that can simultaneously provide a colour video stream, comparable to current webcam technologies, in addition to a depth stream that provides three-dimensional information of the camera’s field of view and is invariant to environmental lighting. An experiment was carried out to characterize the sensor’s accuracy and precision, and to evaluate its performance as an inspection system to determine the orientation of a wheel. Tests were also conducted to determine the effect that changes in the physical environment had on performance. These changes included camera height, lighting and surface material. Results of the experiment have shown that the sensor has an average precision of ±0.12 cm and average accuracy of 0.5 cm, both with less than a 30% change when varying physical features. A discriminant analysis was performed to measure inspection performance, which showed less than 30% change with set separation, but not for set span. No trends were apparent with the change in set span relating to the change in physical features. / Thesis (Master, Mechanical and Materials Engineering) -- Queen's University, 2012-02-29 18:33:20.505

Production downtime caused by machine faults presents a major area of concern for the manufacturing industry and can especially impact the productivity of assembly systems. Traditional fault detection systems use a variety of conventional sensors that measure operating variables such as pressure, force, speed, current and temperature. Faults are detected when a reading from one of these sensors exceeds a preset threshold or does not match the predicted value provided by a mathematical model of the system. The primary disadvantage of these methods is that the relationship between sensor reading and fault is often indirect (if one exists at all). This can lead to time delays between fault occurrence and ‘fault reading’ from a sensor, during which additional machine damage could accumulate. This thesis describes progress with a project whose goal is to examine the effectiveness and feasibility of using machine vision to detect ‘visually cued’ machine faults in automated assembly equipment. It is proposed that machine vision technology could complement traditional methods and improve existing detection systems. Two different vision-based fault detection methods were developed and tests were conducted using a laboratory-scale assembly machine that assembles a simple 3-part component Typical faults that occurred with this machine were targeted for inspection. The first method was developed using Automated Visual Inspection (AVI) techniques that have been used extensively for quality inspection of manufactured products. The LabVIEW 2010 software was used to develop the system. Test results showed that the Colour Inspection tool performed the best with 0% false negative and false positive fault detection rates. Despite some success, this approach was found to be limited as it was unable to detect faults that varied in physical appearance or those that had not been identified prior to testing. The second method was developed using a video event detection method (spatiotemporal volumes) that has previously been used for traffic and pedestrian monitoring. This system was developed with MATLAB software and demonstrated strong false negative and false positive fault detection rates. It also showed the ability to detect faults that had not previously been identified as well as those that varied in appearance. Recommendations were made for future work to further explore these methods. / Thesis (Master, Mechanical and Materials Engineering) -- Queen's University, 2012-07-13 16:04:57.829

An area scan colour camera was used to acquire images of single kernels of Canada Western Red Spring (CWRS) wheat from different growing locations (nine locations in the year 2007, eight locations in the years 2008 and 2009) in Western Canada. Two sample presentation methods were used. In the first method, fifteen kernels from a single location were imaged in a single image and in the second method one kernel from each location were imaged in the same image. Images of individual kernels of barley and rye were also acquired for a classification study. Bulk images of heaped and flat CWRS samples, heaped and flat barley samples, and images of CWRS wheat mixed with different proportion of foreign materials (0%, 2%, 5%, 10%, 20% barley) were acquired. Morphological, colour, and textural features from single kernel images and colour and textural features from bulk grain images were extracted by a program developed by researchers at the Canadian Wheat Board Centre for Grain Storage Research. The top 30 features from the single kernel images of CWRS wheat samples from different growing locations and also different crop years were compared by Scheffe's test. Image features from two types of presentation methods were also compared. Representative of a composite sample which was generated by randomly selecting kernels from each location was compared with individual locations. Three-way classification of CWRS wheat, barley, and rye was done using the top 30 features. For bulk grain image analysis, features from flat bulk grain samples and heaped bulk grain samples were extracted and compared. Image features of CWRS wheat mixed with different percentages of barley were examined, and a cross-validation discriminant classifier was developed to classify CWRS wheat mixed with different percentages of barley. Classifications were also conducted using flat grain as training, flat and heaped grain in testing. Results from this study indicated that most image features from different growing locations and also different crop year samples had significant differences. However, these differences did not influence three-way classification of CWRS wheat, barley, and rye. Features from the composite sample were compared with those from each location. Composite sample features were different from each location. Hence composite samples may not be representative for all locations. However three-way classification using composite sample features gave similar results as in the case of using each location samples. Canada Western Red Spring wheat and barley samples were used in comparing the image features of flat grain and heaped grain. Results indicated that image features from flat grain were different from heaped grain samples. However a two-way classification applied to heaped and flat CWRS wheat, and also heaped and flat barley, gave perfect classification accuracies. Classification models trained using flat grain also gave perfect classification accuracies when tested using flat and heaped grain. A comparison of the top 30 features extracted from images of CWRS wheat mixed with different proportion of barley revealed that grain image features changed after mixing barley. In classification of CWRS wheat mixed with 0, 2, 5, 10, and 20% barley, classification accuracies of 100, 99, 96, 95, and 98% were obtained, respectively.

An area scan colour camera was used to acquire images of single kernels of Canada Western Red Spring (CWRS) wheat from different growing locations (nine locations in the year 2007, eight locations in the years 2008 and 2009) in Western Canada. Two sample presentation methods were used. In the first method, fifteen kernels from a single location were imaged in a single image and in the second method one kernel from each location were imaged in the same image. Images of individual kernels of barley and rye were also acquired for a classification study. Bulk images of heaped and flat CWRS samples, heaped and flat barley samples, and images of CWRS wheat mixed with different proportion of foreign materials (0%, 2%, 5%, 10%, 20% barley) were acquired. Morphological, colour, and textural features from single kernel images and colour and textural features from bulk grain images were extracted by a program developed by researchers at the Canadian Wheat Board Centre for Grain Storage Research. The top 30 features from the single kernel images of CWRS wheat samples from different growing locations and also different crop years were compared by Scheffe's test. Image features from two types of presentation methods were also compared. Representative of a composite sample which was generated by randomly selecting kernels from each location was compared with individual locations. Three-way classification of CWRS wheat, barley, and rye was done using the top 30 features. For bulk grain image analysis, features from flat bulk grain samples and heaped bulk grain samples were extracted and compared. Image features of CWRS wheat mixed with different percentages of barley were examined, and a cross-validation discriminant classifier was developed to classify CWRS wheat mixed with different percentages of barley. Classifications were also conducted using flat grain as training, flat and heaped grain in testing. Results from this study indicated that most image features from different growing locations and also different crop year samples had significant differences. However, these differences did not influence three-way classification of CWRS wheat, barley, and rye. Features from the composite sample were compared with those from each location. Composite sample features were different from each location. Hence composite samples may not be representative for all locations. However three-way classification using composite sample features gave similar results as in the case of using each location samples. Canada Western Red Spring wheat and barley samples were used in comparing the image features of flat grain and heaped grain. Results indicated that image features from flat grain were different from heaped grain samples. However a two-way classification applied to heaped and flat CWRS wheat, and also heaped and flat barley, gave perfect classification accuracies. Classification models trained using flat grain also gave perfect classification accuracies when tested using flat and heaped grain. A comparison of the top 30 features extracted from images of CWRS wheat mixed with different proportion of barley revealed that grain image features changed after mixing barley. In classification of CWRS wheat mixed with 0, 2, 5, 10, and 20% barley, classification accuracies of 100, 99, 96, 95, and 98% were obtained, respectively.