Darrel Greenhill

A multi-agent framework for visual surveillance

We describe an architecture for implementing scene understanding algorithms in the visual surveillance domain. To achieve a high level description of events observed by multiple cameras, many inter-related event-driven processes must be executed. We use the agent paradigm to provide a framework in which these processes can be managed. Each camera has an associated agent, which detects and tracks moving regions of interest. This is used to construct and update object agents. Each camera is calibrated so that image co-ordinates can be transformed into ground plane locations. By comparing properties, two object agents can infer that they have the same referent, i.e. that two cameras are observing the same entity, and as a consequence merge identities. Each objects trajectory is classified with a type of activity, with reference to a ground plane agent. We demonstrate objects simultaneously tracked by two cameras, which infer this shared observation. The combination of the agent framework, and visual surveillance application provides an excellent environment for development and evaluation of scene understanding algorithms.

Adaptive eigen-backgrounds for object detection

Most tracking algorithms detect moving objects by comparing incoming images against a reference frame. Crucially, this reference image must adapt continuously to the current lighting conditions if objects are to be accurately differentiated. In this work, a novel appearance model method is presented based on the eigen-background approach. The image can be efficiently represented by a set of appearance models with few significant dimensions. Rather than accumulating the necessarily enormous training set to generate the eigen model, the described technique builds and adapts the eigen-model online evolving both the parameters and number of significant dimension. For each incoming image, a reference frame may be efficiently hypothesized from a subsample of the incoming pixels. A comparative evaluation that measures segmentation accuracy using large amounts of manually derived ground truth is presented.

Occlusion analysis: Learning and utilising depth maps in object tracking

Complex scenes such as underground stations and malls are composed of static occlusion structures such as walls, entrances, columns, turnstiles and barriers. Unless this occlusion landscape is made explicit such structures can defeat the process of tracking individuals through the scene. This paper describes a method of generating the probability density functions for the depth of the scene at each pixel from a training set of detected blobs, i.e., observations of detected moving people. As the results are necessarily noisy, a regularization process is employed to recover the most self-consistent scene depth structure. An occlusion reasoning framework is proposed to enable object tracking methodologies to make effective use of the recovered depth.

Live cell image analysis of cell-cell interactions reveals the specific targeting of vascular smooth muscle cells by fetal trophoblasts

In early pregnancy, fetal trophoblasts selectively invade and remodel maternal spiral arteries. A healthy pregnancy is dependent on this adaptation to allow sufficient maternal blood to reach the placenta and the developing fetus. However, little is known of the role played by trophoblasts in this adaptation process. In this study, the interactions between trophoblast cells (TC) and vascular smooth muscle cells (VSMC) were examined using novel live cell image analysis methods which allow quantitative assessment of the behaviour of these two cell types in co-culture. TC and VSMC were simultaneously tracked in co-culture and, for each cell type, directionality, speed and the cell-cell interaction were assessed. The overall migratory behaviour of TC was markedly different in the presence of VSMC with co-cultured TC migrating further with directional movement while mono-cultured TC moved more randomly. Furthermore, TC were shown to specifically target VSMC, suggesting that invading TC may initiate targeted vascular remodelling. Analysis of movement behaviour and cell-cell attraction will be useful in other co-culture systems in addition to answering important questions in the reproductive field.

The regulation of trophoblast migration across endothelial cells by low shear stress: consequences for vascular remodelling in pregnancy

AIMS In early human pregnancy placental trophoblasts migrate along uterine spiral arteries (SAs) and remodel these vessels into wide-bore conduits in a process essential for successful pregnancy. Until 10-12 weeks gestation trophoblasts plug spiral arteries, resulting in slow, high-resistance blood flow. This work examined the consequences of these low shear stress conditions on trophoblast migration, adhesion molecule expression, and attraction to chemotactic factors. METHODS AND RESULTS Trophoblasts were cultured on fibronectin or human endothelial cells for 6-12 h under 0.5-6 dyne/cm(2) shear stress using the BioFlux200 system, and imaged by time-lapse microscopy. Computer-based imaging algorithms were developed to automatically quantify migration. Chemotaxis assays were run using parallel flow. Trophoblasts cultured on fibronectin or endothelial cells did not undergo directional migration in 0.5 and 2 dyne/cm(2) cultures; however, in 4 and 6 dyne/cm(2) trophoblasts migrated with the direction of flow (n= 4, P< 0.001). Shear stresses did not affect the speed of trophoblast migration, or adhesion molecule expression (E-selectin, α(4), β(1), and α(v)β(3) integrin). Trophoblasts cultured on endothelial cells migrated into media containing interleukin-8, macrophage chemoattractant protein-1, or Regulated-upon-Activation-Normal-T-cell-Expressed-and-Secreted (RANTES) (n= 5, P< 0.05). CONCLUSIONS Shear stress increases trophoblast migration in the direction of flow, challenging the idea that trophoblasts migrate down spiral arteries retrograde to flow. This suggests that low shear stresses generated by trophoblast plugging of spiral arteries in the first trimester may favour arterial remodelling by preventing the migration with flow seen at higher shear stresses, allowing trophoblasts to migrate down the arteries in response to alternate stimuli such as uterine or endothelial cell-derived chemotactic factors.

Quantitative use of photography in orthognathic outcome assessment

This study reports an independent audit of two aspects of orthognathic surgery, namely control of inter-alar width and mandibular outline asymmetry. Measurements were taken from standardized photographs of a consecutive series of 27 patients, using an on-screen digitizing program (IPTool). All patients had undergone bimaxillary osteotomies involving maxillary impaction and/or advancement, by one surgeon, using a cinch suture for nasal width control. Nine-twelve months after surgery, inter-alar width had increased by just 0.08 cm mean (SD 0.3). Four patients showed an increase of just over 2mm, whilst six showed a small reduction. Based on ratios of size (area) and shape (compactness) of the right and left mandibular segments, there was a small overall improvement in mandibular symmetry (0.019 and 0.005 respectively). Whilst in most of the patients the need for surgery was primarily the correction of antero-posterior and vertical discrepancies, five patients with demonstrable asymmetry showed a clear improvement. In three patients whose asymmetry scores were very mild pre-treatment, there was a small, measured increase in asymmetry, but not to a degree that would be clinically noticeable. At a time when 3D imaging is still unavailable to many clinicians, the results of this study suggest that appropriate measurements taken from carefully standardized conventional photographs can provide a valid and objective means of assessing treatment outcome.

An Agent Society for Scene Interpretation

The paper presents an agent-based framework for use in scene understanding. The framework is suitable for the development of an intelligent distributed system targeted for smart assistant technologies to aid security staff employed to monitor security areas. The framework was used to implement a car park monitoring prototype system.

Segmenting film sequences using active surfaces

Special effects in motion pictures largely employ image compositing to seamlessly join picture elements together from separate sequences. The extraction of objects such as actors from real backgrounds rather than blue screens requires manual tracking of the object boundary by skilled operators: an inaccurate and error-prone task which inevitably generates a bubbling artifact. An active surface model defined over the three dimensional spatio-temporal space is recovered whose intersection with each image plane represents the required object boundary. Such active surfaces are a generalisation of snake contours common in many image processing applications. Initial crude boundaries are drawn in end frames and key intermediate frames to construct an initial surface. A energy minimisation process is used to iteratively refine the location of boundaries in all frames simultaneously. User selected edge data may also be employed to guide the crudely specified initial contours to their correct location.

Rewards and Penalties: A Gamification Approach for Increasing Attendance and Engagement in an Undergraduate Computing Module

This paper describes how a gamified rewards and penalties framework was used to increase attendance and engagement in a level six undergraduate computing module teaching game production. The framework was applied to the same module over two consecutive years: a control year and a trial year. In both years the tutor, assignments and assessment strategies were the same and daily attendance was recorded. In the module, students work in multi-disciplinary teams to complete an assignment to build a computer game prototype. Unequal contribution to team projects by other students is a frequently voiced complaint to lecturers setting team assignments: a problem which is only partially solved by peer assessments, which are a retrospective analysis. The gamification framework provides a method for the lecturer to quickly identify disengaging students and to re-motivate them. Partnership between student and teacher, both parties must present themselves in order for that exchange of knowledge to take place. If unequal team contribution is a constant problem for students, then empty lecture halls can be considered similarly difficult for educators. This paper addresses three key points: 1) Does the rewards/penalties framework improve attendance? 2) If yes, does improved attendance result in improved assessments? 3) Does the framework improve engagement and performance in student teams? This paper presents quantitative evidence to answer the first two and offers speculative comments on the third. Initial results suggest that the rewards and penalties framework improves attendance and increases student performance and overall grade. Speculatively, the framework appears to be effective in increasing motivation. Informal student commentary indicates that while motivation is not improved across the cohort, those that are motivated contribute significantly more time and effort to the project. Rewards proved successful in improving completion of previously resisted tasks and in attracting students to attend classes they would otherwise miss.

QUERY MELTING - A New Paradigm for GIS Multiple Query Optimization

Recently, non-expert mobile-user applications have been developed to query Geographic Information Systems (GIS) particularly Location Based Services where users ask questions related to their position whether they are moving (dynamic) or not (static). A new Iconic Visual Query Language (IVQL) has been developed to handle proximity analysis queries that find k-nearest-neighbours and objects within a buffer area. Each operator in IVQL queries corresponds to an execution plan to be evaluated by the GIS server. Since commonalities exist between the execution plans, the same operations are executed many times leading to slow results. Hence, the need arises to develop a multi-user dynamic complex query optimizer that handles commonalities and processes the queries faster especially with the large-scale of mobile-users. We present a new query processor, a generic optimization framework for GIS and a middleware, which employs the new Query Melting paradigm (QM) that is based on the sharing paradigm and push-down optimization strategy. QM is implemented through a new Melting-Ruler strategy that works at the lowlevel, melts repetitions in plans to share spatial areas, temporal intervals, objects, intermediate results, maps, user locations, and functions, then re-orders them to get time-cost effective results, and is illustrated using a sample tourist GIS system.