Lecture videos are good sources for people to learn new things. Students commonly use online videos to explore various domains. However, some recorded videos are posted on online platforms without being post-processed due to technology and resource limitations. In this work, we focus on the research of developing an intelligent system to automatically extract essential information, including the main instructor and screen, in a lecture video in several scenarios by using modern deep learning techniques. This thesis aims to combine the extracted essential information to render the videos and generate a new layout with a smaller file size than the original one. Another benefit of using this approach is that the users may save video post-processing time and costs. State-of-the-art object detection models, an algorithm to correct screen display, tracking the instructor, and other deep learning techniques were adopted in the system to detect both the main instructor and the screen in given videos without much of the computational burden. There are four main contributions: 1. We built an intelligent video analysis and post-processing system to extract and reframe detected objects from lecture videos. 2. We proposed a post-processing algorithm to localize the frontal human torso position in processing a sequence of frames in the videos. 3. We proposed a novel deep learning approach to distinguish the main instructor from other instructors or audiences in several complex situations. 4. We proposed an algorithm to extract the four edge points of a screen at the pixel level and correct the screen display in various scenarios.

With the exponential growth of the digital data, video content analysis (e.g., action, event recognition) has been drawing increasing attention from computer vision researchers. Effective modeling of the objects, scenes, and motions is critical for visual understanding. Recently there has been a growing interest in the bio-inspired deep learning models, which has shown impressive results in speech and object recognition. The deep learning models are formed by the composition of multiple non-linear transformations of the data, with the goal of yielding more abstract and ultimately more useful representations. The advantages of the deep models are three fold: 1) They learn the features directly from the raw signal in contrast to the hand-designed features. 2) The learning can be unsupervised, which is suitable for large data where labeling all the data is expensive and unpractical. 3) They learn a hierarchy of features one level at a time and the layerwise stacking of feature extraction, this often yields better representations. However, not many deep learning models have been proposed to solve the problems in video analysis, especially videos “in a wild”. Most of them are either dealing with simple datasets, or limited to the low-level local spatial-temporal feature descriptors for action recognition. Moreover, as the learning algorithms are unsupervised, the learned features preserve generative properties rather than the discriminative ones which are more favorable in the classification tasks. In this context, the thesis makes two major contributions. First, we propose several formulations and extensions of deep learning methods which learn hierarchical representations for three challenging video analysis tasks, including complex event recognition, object detection in videos and measuring action similarity. The proposed methods are extensively demonstrated for each work on the state-of-the-art challenging datasets. Besides learning the low-level local features, higher level representations are further designed to be learned in the context of applications. The data-driven concept representations and sparse representation of the events are learned for complex event recognition; the representations for object body parts iii and structures are learned for object detection in videos; and the relational motion features and similarity metrics between video pairs are learned simultaneously for action verification. Second, in order to learn discriminative and compact features, we propose a new feature learning method using a deep neural network based on auto encoders. It differs from the existing unsupervised feature learning methods in two ways: first it optimizes both discriminative and generative properties of the features simultaneously, which gives our features a better discriminative ability. Second, our learned features are more compact, while the unsupervised feature learning methods usually learn a redundant set of over-complete features. Extensive experiments with quantitative and qualitative results on the tasks of human detection and action verification demonstrate the superiority of our proposed models.

Crew race strategy is typically formulated by coaches based on rowing tradition and years of experience. However, coaching strategies are not generally supported by empirical evidence and decision-support models. Previous models of crew race strategy have been constrained by the sparse information published on crew race performance (quarterly 500-meter splits). Empirical research has merely summarized which quarterly splits averaged the fastest and slowest relative to the other splits and relative to the average speed of the other competitors. Video records of crew race world championships provide a rich source of data for those capable and patient enough to mine this level of detail. This dissertation is based on a precise frame-by-frame video analysis of five world championship rowing finals. With six competing crews per race, a database of 75 race-pair duels was compiled that summarizes race positioning, competitive drives, and relative stroke rates at 10-meter intervals recorded with photo-finish precision (30 frames per second). The drive-based research pioneered in this dissertation makes several contributions to understanding the dynamics of crew race strategy and performance: 1) An 8-factor conceptual model of crew race performance. 2) A generic drive model that decomposes how pairs of crews duel in a race. 3) Graphical summaries of the rates and locations of successful and unsuccessful drives. 4) Contour lines of the margins that winning crews hold over the course of the race. 5) Trend lines for what constitutes a probabilistically decisive lead as a function of position along the course, seconds behind the leader, and whether the trailing crew is driving. This research defines a new drive-based vocabulary for evaluating crew race performance for use by coaches, competitors and race analysts. The research graphically illustrates situational parameters helpful in formulating race strategy and guiding real-time decision-making by competitors. This research also lays the foundation for future industrial engineering decision-support models and associated parameters as applied to race strategy and tactics.

Providing effective supervision of student teachers and interns is critical in preparing quality special education teachers. To decrease the time commitment of supervisors, researchers have suggested using virtual observations which are generally viewed as a valuable resource. This study examined the experience of teacher candidates supervised with a combination of face-to-face visits and video observations. Groups of university faculty (supervisors) and students (teacher candidates) from a western university participated with both methods of supervision over three years and provided feedback to researchers. Results highlighted participants satisfaction with the supervision process, including advantages and concerns with each type and ways the combination of methods gave added value and efficiency. Benefits of using the combination of observation types, which was most valued by both supervisors and candidates, were convenience and flexibility of scheduling, opportunities for self-evaluation, and the nature and timeliness of feedback. Future research might focus on using the combination of supervision methods with licensure candidates in other groups of professionals and teachers or examine specific aspects of using the technology.

Over the past decade, there has been an explosion in the volume of video data, including internet videos and surveillance camera footage. These videos often feature extended durations with unedited content, predominantly filled with background clutter, while the relevant activities of interest occupy only a small portion of the footage. Consequently, there is a compelling need for advanced processing techniques to automatically analyze this vast reservoir of video data, specifically with the goal of identifying the segments that contain the events of interest. Given that humans are the primary subjects in these videos, comprehending human activities plays a pivotal role in automated video analysis. This thesis seeks to tackle the challenge of detecting human activities from untrimmed videos, aiming to classify and pinpoint these activities both in their spatial and temporal dimensions. To achieve this, we propose a modular approach. We begin by developing a temporal activity detection framework, and then progressively extend the framework to support activity detection in the spatio-temporal dimension. To perform temporal activity detection, we introduce an end-to-end trainable deep learning model leveraging 3D convolutions. Additionally, we propose a novel and adaptable fusion strategy to combine both the appearance and motion information extracted from a video, using RGB and optical flow frames. Importantly, we incorporate the learning of this fusion strategy into the activity detection framework. Building upon the temporal activity detection framework, we extend it by incorporating a spatial localization module to enable activity detection both in space and time in a holistic end-to-end manner. To accomplish this, we leverage shared spatio-temporal feature maps to jointly optimize both spatial and temporal localization of activities, thus making the entire pipeline more effective and efficient. Finally, we introduce several novel techniques for modeling actor motion, specifically designed for efficient activity recognition. This is achieved by harnessing 2D pose information extracted from video frames and then representing human motion through bone movement, bone orientation, and body joint positions. Our experimental evaluations, conducted using benchmark datasets, showcase the effectiveness of the proposed temporal and spatio-temporal activity detection methods when compared to the current state-of-the-art methods. Moreover, the proposed motion representations excel in both performance and computational efficiency. Ultimately, this research shall pave the way forward towards imbuing computers with social visual intelligence, enabling them to comprehend human activities in any given time and space, opening up exciting possibilities for the future.

The aim of this study is to investigate which effects the introduction of the new legislation for bicyclestreets has on road users. The focus has been on studying road users' compliance, accessibility andexperience of bicycle streets. Since the legislation came into force in December 2020, vehicles entering abicycle street have a duty to give way to road users on the bicycle street and drivers of motor vehicleshave to adapt their speed to the bicycle traffic. There is also a general speed limit of 30 km/h and parkingis only allowed in designated areas.The report consists of three main sections, the first of which reviews the history and dialogue behind thecurrent legislation and addresses some of the main points of criticisms leveled at it. This was exploredthrough a detailed review of the available documents and studies dealing with bicycle streets. Thesecond part aims to review the conditions for cycle streets in Sweden and to provide an overview of theopportunities and challenges that exist. This part includes, among other things, thorough examinationsof previous studies of bicycle streets, existing bicycle streets in Sweden, similar legislation and aninternational outlook where bicycle streets in other countries in Northern Europe are studied. The lastpart is a case study of one of Sweden's first bicycle streets in Varberg where, among other things,speeds, movement patterns and yielding behavior of road users are studied. This was investigatedthrough radar measurements, video analysis and visual observations on site. Furthermore, two differentsurveys were carried out, one of which aimed to find out the opinions and experiences of road users onthe bicycle street in Varberg. The second survey was of a more general nature and aimed to collect roadusers' opinions on issues related to bicycle streets.The results indicate that there is a general consensus among experts, planners and road users thatbicycle streets have the potential to improve cyclists' accessibility and road safety and, in the long term,to encourage more people to cycle. However, several shortcomings are found in the existing legislationthat many believe prevent the bicycle street from achieving its purpose. For example, there are currentlyno official guidelines on how a bicycle street should be designed and there is also older legislation thatprevents cyclists from being able to use the whole roadway and cycle side by side. The case study revealsseveral interesting findings that confirm several points of criticism of bicycle streets and its legislation.Among other things, the majority of the observed cyclists chose to take up space in the roadway and alsoride side by side. The survey results also indicate that a change in traffic regulations to allow this onbicycle streets is something that is desired by a majority. Based on the results of the study, severalsuggestions are made for regulation changes that would make the bicycle street better serve its purpose.For example, it is suggested that cyclists should always be allowed to use the whole roadway and rideside by side on bicycle streets. It is also recommended that the responsible authorities develop nationalrecommendations for the physical design of cycle lanes as soon as possible. As for other common viewson regulatory changes, such as whether or not overtaking of cyclists should be allowed and what speedlimit should apply on bicycle streets, further studies need to be carried out before any conclusions canbe drawn.

No description available.

Concussions occur frequently as a result of tackles in ice hockey. Analysis of video material may provide an understanding of the relationship between the kinematics of collisions and the risk for injury. In this thesis, two video analysis methods were used to estimate the impact velocities of 22 ice hockey tackles that resulted in concussions. The Point tracking method uses tracking of user-defined object points on the players and ice to estimate the velocities. It was used in an earlier thesis. A deep learning-based method was implemented in this thesis. It uses a pre-trained deep learning model to detect the players in each frame of the video. Both methods were validated in this thesis using soccer videos containing accelerometer data from the players. The mean error was 25.6 % for the Point tracking method and 43.1 % for the Deep learning method. The difference was not significant. Both methods calculate the player velocity as a mean from a given number of video frames before impact. The choice of the number of frames did not significantly affect the difference in estimated velocities between the Point tracking method and the Deep learning method. The Point tracking method succeeded in estimating velocities in 17 cases. The mean velocities for the attacking and injured players were 10.5 m/s and 9.3 m/s, respectively. The Deep learning method succeeded in 9 cases, and the mean velocities were 9.7 m/s and 9.5 m/s. The velocities are higher than what has been found in earlier research, suggesting that both methods may be biased towards estimating too high velocities. More investigation needs to be done to evaluate the methods’ performance, possibly by comparing with accelerometer data from ice hockey.

This dissertation thesis describes the area of multimedia semantics which is a research area that brings together research streams that until recently run separately. The aim of the work is to provide an insight to all areas from this wide discipline and give an outlook on current problems especially to the semantic gab phenomena. Number of findings and outcomes in this work comes from international project K-Space, in which the author took part for three years. The extensive theoretical introduction into problematic is followed by a list state-of-the-art application from this area and overview of KIZI activities and involvements in the European project. The contribution of the work is a research on textual resources complementary to video and experiments with automatic detection of sporting events based on pre-classified examples and trained model. The practical contribution is also a demo web application that shows all the resources together and allows non-linear browsing of events.

Esta dissertação apresenta um novo método de segmentação de movimento baseado na obtenção dos contornos e em filtros morfológicos. A nova técnica apresenta vantagens em relação ao número de falsos positivos e falsos negativos em situações específicas quando comparada às técnicas tradicionais. / This work presents a novel motion segmentation technique based in contours and in morphological filters. It presents advantages in the number of false positives and false negatives in some situations when compared to the classic techniques.