Browsing by Subject "Computer vision"

Now showing 1 - 17 of 17

A Comprehensive System for Assessing Truck Parking Availability
(Center for Transportation Studies, University of Minnesota, 2017-01) Morris, Ted; Murray, Dan; Fender, Kate; Weber, Amanda; Morellas, Vassilios; Cook, Doug; Papanikolopoulos, Nikos
Commercial heavy vehicle (CHV) drivers are required under federal Hours of Services (HOS) rules to rest and take breaks to reduce driving while fatigued. CHV drivers and operators must balance compliance to the HOS rules against on-time delivery requirements as well as shorter lead times to plan their trips, thereby making location and parking availability of rest area facilities more critical. Without timely, accurate parking availability information, drivers are left with the dilemma of continuing to drive fatigued, drive beyond HOS CHV operation limits, or park illegally on highway shoulders or ramps—all potential safety hazards. In this study, a multi-view camera system was designed and evaluated to detect truck parking space occupancy in real-time through extensive field operational testing. A system architecture was then developed to disseminate up-to-the-minute truck parking information through three separate information delivery systems: 1) Roadside Changeable Message Signs (CMS), 2) Internet/Website information portal, and 3) an onboard geolocation application. The latter application informs the driver of parking availability of one or more parking facilities that are downstream from their current direction of travel. All three notification mechanisms were evaluated during the field test. Survey studies were conducted to provide feedback from commercial heavy vehicle drivers and operators to better understand their perceptions of parking shortages and utility of the parking information delivery mechanisms. Overall, the system has proven to provide 24/7 around-the-clock per-space parking status with no need for manual interventions to correct detection errors, with per parking space accuracy typically equal to or exceeding 95 percent. The concept of operations field tests demonstrated the feasibility of the technical approach and the potential to alter freight borne trip behaviors by allowing drivers and carriers to plan stops and improve trip efficiency.
Computer vision applications of head motion and cell analysis
(2013-05) Hashemi, Jordan K.
Computer vision approaches have been applied to a wide variety of domains. These approaches are able to replicate human processes as well as provide new insights. Once developed, computer vision approaches also permit high throughput analysis of processes that would otherwise be tedious. In this thesis, we focus on developing tools that replicate and go above human analysis for two applications. We aim at developing tools to track infants` head motions in a non-intrusive manner during an autism assessment. We propose a method to track facial features from a single camera and use their changing locations to sequentially update the motions. We also aim at developing an elliptical-cell and multiple-fluorescence segmentation program to allow for the high throughput analysis of C.albicans and S.cerevisiae yeast cultures. We propose to segment pseudohyphal cells based on an extension of the circular Hough transform; as well as, segment multiple types of fluorescence labeling.
Computer Vision Methods to Characterize the Morphology of Mouse Skulls for Neuroscience Applications
(2023-02) Gulner, Beatrice
Computer vision is a powerful tool for automating the characterization of biological specimen morphology. Classical morphometric studies have provided crucial insights into the skull anatomy of commonly used wildtype (WT) laboratory mice strains such as the C57BL/6. With the increasing use of transgenic (TG) animals in neuroscience research, it is important to determine whether the results from morphometric studies performed on WT strains can be extended to TG strains derived from these WT strains. In this thesis, we first report a new computer vision-based analysis pipeline for surveying mouse skull morphology using Microcomputed Tomography (µCT) scans. We applied this pipeline to study and compare eight cohorts of adult mice from two strains, including both male and female mice at two age points. We found that the overall skull morphology was generally conserved between cohorts, with only 13% of landmark distance differences reaching statistical significance. In addition, we surveyed the dorsal skull bone thickness differences between cohorts. We observed significantly thicker dorsal, parietal, and/or interparietal bones in WT, male, or older mice for 53% of thickness comparisons. Many neuroscience experiments require penetrating the mouse skull to record or modulate neural activity in the brain. Craniotomy procedures on sub-millimeter thick skull tissue are time-consuming to perform manually and require substantial training to attain an acceptable success rate. Previous researchers have used automation to reduce the training needed, increase speed, and minimize variability, but insufficient knowledge of the dorsal skull thickness limits their performance. We thus present a fast, non-invasive method which employs preoperative Optical Coherence Tomography (OCT) imaging to guide a robot to perform single-pass craniotomies in mice. The mouse skull is scanned with an OCT scanner immediately prior to surgery, then a custom computer vision-based analysis pipeline extracts an approximately 10 µm axial by 20 µm lateral resolution 3D profile of the dorsal and ventral surfaces of the mouse calvaria within the Field of View (FOV). A cutting path is generated based on the depth of the ventral surface along the desired craniotomy path. Comparison with µCT skull thickness data and preliminary surgery results indicates that this method provides an acceptable profile across most of the mouse dorsal skull, though more iteration is required to ensure accurate measurement of the area around the lambdoid sinus.
Counting Empty Parking Spots at Truck Stops Using Computer Vision
(Center for Transportation Studies, 2011-05) Pushkar, Modi; Vassilios, Morellas; Papanikolopoulos, Nikolaos
For at least the past decade, truck driver fatigue has been thought to be a contributing factor in a number of heavy truck accidents. For better utilization of truck stops and to provide truck drivers with safe rest options, we are designing an automated truck stop management system that can compute occupancy rates at stops and notify drivers about the availability of parking spots using variable message displays located about 30 or 40 miles before the stop. Our system detects, classifies and localizes vehicles on the truck stop's grounds by using a set of video cameras, from which video frames are analyzed in real-time.
Data Mining of Traffic Video Sequences
(University of Minnesota Center for Transportation Studies, 2009-09) Joshi, Ajay J.; Papanikolopoulos, Nikolaos
Automatically analyzing video data is extremely important for applications such as monitoring and data collection in transportation scenarios. Machine learning techniques are often employed in order to achieve these goals of mining traffic video to find interesting events. Typically, learning-based methods require significant amount of training data provided via human annotation. For instance, in order to provide training, a user can give the system images of a certain vehicle along with its respective annotation. The system then learns how to identify vehicles in the future - however, such systems usually need large amounts of training data and thereby cumbersome human effort. In this research, we propose a method for active learning in which the system interactively queries the human for annotation on the most informative instances. In this way, learning can be accomplished with lesser user effort without compromising performance. Our system is also efficient computationally, thus being feasible in real data mining tasks for traffic video sequences.
Deployment of Practical Methods for Counting Bicycle and Pedestrian Use of a Transportation Facility
(Intelligent Transportation Systems Institute, Center for Transportation Studies, 2012-01) Somasundaram, Guruprasad; Morellas, Vassilios; Papanikolopoulos, Nikolaos
The classification problem of distinguishing bicycles from pedestrians for traffic counting applications is the objective of this research project. The scenes that are typically involved are bicycle trails, bridges, and bicycle lanes. These locations have heavy traffic of mainly pedestrians and bicyclists. A vision-based system overcomes many of the shortcomings of existing technologies such as loop counters, buried pressure pads, infra-red counters, etc. These methods do not have distinctive profiles for bicycles and pedestrians. Also most of these technologies require expert installation and maintenance. Cameras are inexpensive and abundant and are relatively easy to use, but they tend to be useful as a counting system only when accompanied by powerful algorithms that analyze the images. We employ state-of-the-art algorithms for performing object classification to solve the problem of distinguishing bicyclists from pedestrians. We detail the challenges that are involved in this particular problem, and we propose solutions to address these challenges. We explore common approaches of global image analysis aided by motion information and compare the results with local image analysis in which we attempt to distinguish the individual parts of the composite object. We compare the classification accuracies of both approaches on real data and present detailed discussion on practical deployment factors.
Development of a Sensor Platform for Roadway Mapping: Part B – Mapping the Road Fog Lines
(Minnesota Department of Transportation, 2015-04) Davis, Brian; Donath, Max
Our objective is the development and evaluation of a low-cost, vehicle-mounted sensor suite capable of generating map data with lane and road boundary information accurate to the 10 cm (4 in) level. Such a map could be used for a number of different applications including GNSS/GPS based lane departure avoidance systems, smart phone based dynamic curve speed warning systems, basemap improvements, among others. The sensor suite used consists of a high accuracy GNSS receiver, a side-facing video camera, and a computer. Including cabling and mounting hardware, the equipment costs were roughly $30,000. Here, the side-facing camera is used to record video of the ground adjacent to the passenger side of the vehicle. The video is processed using a computer vision algorithm that locates the fog line within the video frame. Using vehicle position data (provided by GNSS) and previously collected video calibration data, the fog line is located in real-world coordinates. The system was tested on two roads (primarily two-lane, undivided highway) for which high accuracy (<10 cm) maps were available. This offset between the reference data and the computed fog line position was generally better than 7.5 cm (3 in). The results of this work demonstrate that it is feasible to use a camera to detect the position of a road’s fog lines, or more broadly any other lane markings, which when integrated into a larger mobile data collection system, can provide accurate lane and road boundary information about road geometry.
A fast, low-cost, computer-vision based approach for tracking surgical tools
(2013-08) Dockter, Rodney Lee
The number of Robot-Assisted Minimally Invasive Surgery (RMIS) procedures has grown immensely in recent years. Like Minimally Invasive Surgeries (MIS), RMIS procedures provide improved patient recovery time and reduced trauma due to smaller incisions relative to traditional open procedures. Given the rise in RMIS procedures, several organizations and companies have made efforts to develop training tasks and certification criteria for the da Vinci robot. Each training task is evaluated with various quantitative criteria such as completion time, total tool path length and economy of motion, which is a measurement of deviation from an `ideal' path. All of these metrics can benefit greatly from an accurate, inexpensive and modular tool tracking system that requires no modification to the existing robot. While the da Vinci uses joint kinematics to calculate the tool tip position and movement internally, this data is not openly available to users. Even if this data was open to researchers, the accuracy of kinematic calculations of end effector position suffers from compliance in the joints and links of the robot as well as finite uncertainties in the sensors. In order to and an accurate, available and low-cost alternative to tool tip localization, we have developed a computer vision based design for surgical tool tracking. Vision systems have the added benefit of being relatively low cost with typical high resolution webcams costing around 50 dollars. We employ a joint geometric constraint - Hough transform method for locating the tool shaft and subsequently the tool tip. The tool tracking algorithm presented was evaluated on both an experimental webcam setup as well as a da Vinci Endoscope used in real surgeries. This system can accurately locate the tip of a robotic surgical tool in real time with no augmentation of the tool. The proposed algorithm was evaluated in terms of speed and accuracy. This method achieves an average 3D positional tracking accuracy of 3.05 mm and at 25.86 frames per second for the experimental webcam setup. For the da Vinci endoscope setup, this solution achieves a frame rate of 26.99 FPS with an average tracking accuracy of 8.68 mm in 3D and 11.88 mm in 2D. The system demonstrated successful tracking of RMIS tools from captured video of a real patient case.
Image Enhancement Algorithms to Improve Robustness and Accuracy of Pre-trained Neural Networks for Autonomous Driving
(2023-01) Joshi, Himanshu
This research proposes a generalized data-driven approach for improving the pre-training image datasets fed to neural networks(NNs). The algorithms developed and tested in this work could substantially enhance the image bringing out the critical spatial information in the image for better NN performance. This image enhancement technique consists of two main components: image colorization and contrast enhancement. Image colorization is implemented to obtain a color-corrected image from a grayscale image. The traditional global contrast enhancement algorithm is extended to Smoothened Variable Local Dynamic Contrast Improvement (SVL-DCI) to boost local contrasts within an image frame that suffers from under/over-exposed lighting conditions. SVL-DCI algorithm is developed and thoroughly tested in the present thesis that could run in real-time as a pre-training algorithm for NNs. We implemented SVL-DCI on the 3P lab dataset of 2470 images and observed competitive improvement in the performance of the investigated NNs for object recognition and lane detection.
Multi-Camera Monitoring of Human Activities at Critical Transportation Infrastructure Sites
(University of Minnesota Center for Transportation Studies, 2008-06) Ribnick, Evan; Joshi, Ajay J.; Papanikolopoulos, Nikolaos P.
The goal of this work is to provide a system which can aid in monitoring crowded urban environments, which often contain tight groups of people. In this report, we consider the problem of counting the number of people in the scene and also tracking them reliably. We propose a novel method for detecting and estimating the count of people in groups, dense or otherwise, as well as tracking them. Using prior knowledge obtained from the scene and accurate camera calibration, the system learns the parameters required for estimation. This information can then be used to estimate the count of people in the scene, in real time. Groups are tracked in the same manner as individuals, using Kalman filtering techniques. Favorable results are shown for groups of various sizes moving in an unconstrained fashion.
Portable Traffic Data Processor
(University of Minnesota Center for Transportation Studies, 2008-09) Papanikolopoulos, Nikoloas; Veeraraghavan, Harini
Automatic extraction of events from video sequences has important applications in a variety of Intelligent Transportation Systems (ITS) problems including scene monitoring, traffic data collection, intersection monitoring, etc. When deploying a system that recognizes events automatically from video sequences, two important things to consider are the real-time analysis of the video sequences and fast learning times required for learning the different classes of events in a scene. A related requirement which is often ignored is the limited reliance of the learning system on the user provided knowledge. In this work, we present an innovative technique for detecting the different events in video sequences through a semi-supervised learning method.
Practical Methods for Analyzing Pedestrian and Bicycle Use of a Transportation Facility
(Minnesota Department of Transportation Office of Research Services, 2010-02) Somasundaram, Guruprasad; Morellas, Vassilios; Papanikolopoulos, Nikolaos P.
The objective of the project is to analyze existing technologies used for the process of generating counts of bicycles and pedestrians in transportation facilities such as walk and bicycle bridges, urban bicycle routes, bicycle trails etc. The advantages and disadvantages of each existing technology which is being applied to counting has been analyzed and some commercially available products were listed. A technical description of different methods that were considered for vision based object recognition is also mentioned along with the reasons as to why such methods were overlooked for our problem. Support Vector Machines were used for classification based on a vocabulary of features built using interest point detectors. After finalizing the software and hardware, five sites were picked for filming and about 10 hours of video was acquired in all. A portion of the video data was used for training and the remainder was used for testing the algorithm’s accuracy. Results of counts are provided and an interpretation of these results is provided in this report. Upon detailed analysis the reasons for false counts and undercounting in some cases have been identified and current work concerns dealing with these issues. Changes are being made to the system to improve the accuracy with the current level of training and make the system available for practitioners to perform counting.
A Real-Time Truck Availability System for the State of Wisconsin
(Center for Transportation Studies, University of Minnesota, 2018-05) Morris, Ted; Henderson, Travis; Morellas, Vassilios; Papanikolopoulos, Nikos
Independent of truck parking capacity shortages, obtaining reliable and timely information has been receiving considerable attention nationally as of late. The situation has been exacerbated by increasing levels of freightborne truck volumes along many regional and interstate corridors and the need for carriers and drivers to balance efficient transport with required periods to park and rest to minimize driver fatigue. Interstate 94, a nationally designated freight corridor, as it passes through the Upper Midwest, including Wisconsin, shares this problem. A multi-camera computer vision detection system was deployed at a state sponsored rest area truck parking facility 67 miles east of Minneapolis. A key aspect of the system is that it is a completely automated 24/7, non-intrusive, parking detection system; there is no need to intervene with manual resets or re-calibration procedures, and pavements are not disturbed. Secondly, a region-wide truck parking notification architecture, recognized as an emerging national standard, was integrated with the detection system to provide real-time roadside truck parking notifications upstream of the facilities, as well as notification to other third party stakeholders. The overall detection accuracy was between 90 and 95 percent during up-to-the minute, per-space parking status notifications.
Robust Group Synchronization via Quadratic Programming
(2023-06) Wyeth, Cole
We review existing methods for the group synchronization problem and discuss ournovel quadratic programming formulation for estimating the corruption levels in group synchronization, and use these estimates to solve this problem. Our objective function exploits the cycle consistency of the group and we thus refer to our method as detection and estimation of structural consistency (DESC). This general framework can be extended to other algebraic and geometric structures. Our formulation has the following advantages: it can tolerate corruption as high as the information-theoretic bound, it does not require a good initialization for the estimates of group elements, it has a simple interpretation, and under some mild conditions the global minimum of our objective function exactly recovers the corruption levels. We demonstrate the competitive accuracy of our approach on both synthetic and real data experiments of rotation averaging.
Sparse models for positive definite matrices
(2015-02) Sivalingam, Ravishankar
Sparse models have proven to be extremely successful in image processing, computer vision and machine learning. However, a majority of the effort has been focused on vector-valued signals. Higher-order signals like matrices are usually vectorized as a pre-processing step, and treated like vectors thereafter for sparse modeling. Symmetric positive definite (SPD) matrices arise in probability and statistics and the many domains built upon them. In computer vision, a certain type of feature descriptor called the region covariance descriptor, used to characterize an object or image region, belongs to this class of matrices. Region covariances are immensely popular in object detection, tracking, and classification. Human detection and recognition, texture classification, face recognition, and action recognition are some of the problems tackled using this powerful class of descriptors. They have also caught on as useful features for speech processing and recognition.Due to the popularity of sparse modeling in the vector domain, it is enticing to apply sparse representation techniques to SPD matrices as well. However, SPD matrices cannot be directly vectorized for sparse modeling, since their implicit structure is lost in the process, and the resulting vectors do not adhere to the positive definite manifold geometry. Therefore, to extend the benefits of sparse modeling to the space of positive definite matrices, we must develop dedicated sparse algorithms that respect the positive definite structure and the geometry of the manifold. The primary goal of this thesis is to develop sparse modeling techniques for symmetric positive definite matrices. First, we propose a novel sparse coding technique for representing SPD matrices using sparse linear combinations of a dictionary of atomic SPD matrices. Next, we present a dictionary learning approach wherein these atoms are themselves learned from the given data, in a task-driven manner. The sparse coding and dictionary learning approaches are then specialized to the case of rank-1 positive semi-definite matrices. A discriminative dictionary learning approach from vector sparse modeling is extended to the scenario of positive definite dictionaries. We present efficient algorithms and implementations, with practical applications in image processing and computer vision for the proposed techniques.
Structured sparse models for classification
(2012-11) Castrodad, Alexey
The main focus of this thesis is the modeling and classification of high dimensional data using structured sparsity. Sparse models, where data is assumed to be well represented as a linear combination of a few elements from a dictionary, have gained considerable attention in recent years, and its use has led to state-of-the-art results in many signal and image processing tasks. The success of sparse modeling is highly due to its ability to efficiently use the redundancy of the data and find its underlying structure. On a classification setting, we capitalize on this advantage to properly model and separate the structure of the classes. We design and validate modeling solutions to challenging problems arising in computer vision and remote sensing. We propose both supervised and unsupervised schemes for the modeling of human actions from motion imagery under a wide variety of acquisition conditions. In the supervised case, the main goal is to classify the human actions in the video given a predefined set of actions to learn from. In the unsupervised case, the main goal is to analyze the spatio-temporal dynamics of the individuals in the scene without having any prior information on the actions themselves. We also propose a model for remotely sensed hysperspectral imagery, where the main goal is to perform automatic spectral source separation and mapping at the subpixel level. Finally, we present a sparse model for sensor fusion to exploit the common structure and enforce collaboration of hyperspectral with LiDAR data for better mapping capabilities. In all these scenarios, we demonstrate that these data can be expressed as a combination of atoms from a class-structured dictionary. These data representation becomes essentially a “mixture of classes,” and by directly exploiting the sparse codes, one can attain highly accurate classification performance with relatively unsophisticated classifiers.
Video Detection and Classification of Pedestrian Events at Roundabouts and Crosswalks
(Intelligent Transportation Systems Institute, Center for Transportation Studies, 2013-08) Morris, Ted; Li, Xinyan; Morellas, Vassilios; Papanikolopoulos, Nikos
A well-established technique for studying pedestrian safety is based on reducing data from video-based in-situ observation. The extraction and cataloging from recorded video of pedestrian crossing events has largely been achieved manually. Although the manual methods are generally reliable, they are extremely time-consuming. As a result, more detailed, encompassing site studies are not practical unless the mining for these events can be automated. The study investigated such a tool based on utilizing a novel image processing algorithm recently developed for the extraction of human activities in complex scenes. No human intervention other than defining regions of interest for approaching vehicles and the pedestrian crossing areas was required. The output quantified general event indicators—such as pedestrian wait time, and crossing time and vehicle-pedestrian yield behaviors. Such data can then be used to guide more detailed analyses of the events to study potential vehicle-pedestrian conflicts and their causal effects. The evaluation was done using an extensive set of multi-camera video recordings collected at roundabouts. The tool can be used to support other pedestrian safety research where extracting potential pedestrian-vehicle conflicts from video are required, for example at crosswalks at urban signalized and uncontrolled intersections.

University Digital Conservancy

Browse by Subject

Browsing by Subject "Computer vision"