Jingxin Xu

Dynamic texture reconstruction from sparse codes for unusual event detection in crowded scenes

Unusual event detection in crowded scenes remains challenging because of the diversity of events and noise. In this paper, we present a novel approach for unusual event detection via sparse reconstruction of dynamic textures over an overcomplete basis set, with the dynamic texture described by local binary patterns from three orthogonal planes (LBPTOP). The overcomplete basis set is learnt from the training data where only the normal items observed. In the detection process, given a new observation, we compute the sparsecoefficients using the Dantzig Selector algorithm which was proposed in the literature of compressed sensing. Then the reconstruction errors are computed, based on which we detect the abnormal items. Our application can be used to detect both local and global abnormal events. We evaluate our algorithm on UCSD Abnormality Datasets for local anomaly detection, which is shown to outperform current state-of-the-art approaches, and we also get promising results for rapid escape detection using the PETS2009 dataset.

Unusual Scene Detection Using Distributed Behaviour Model and Sparse Representation

The ability to detect unusual events in surviellance footage as they happen is a highly desireable feature for a surveillance system. However, this problem remains challenging in crowded scenes due to occlusions and the clustering of people. In this paper, we propose using the Distributed Behavior Model (DBM), which has been widely used in computer graphics, for video event detection. Our approach does not rely on object tracking, and is robust to camera movements. We use sparse coding for classification, and test our approach on various datasets. Our proposed approach outperforms a state-of-the-art work which uses the social force model and Latent Dirichlet Allocation.

ProcessProfiler3D:a visualisation framework for log-based process performance comparison

An organisation can significantly improve its performance by observing how their business operations are currently being carried out. A great way to derive evidence-based process improvement insights is to compare the behaviour and performance of processes for different process cohorts by utilising the information recorded in event logs. A process cohort is a coherent group of process instances that has one or more shared characteristics. Such process performance comparisons can highlight positive or negative variations that can be evident in a particular cohort, thus enabling a tailored approach to process improvement. Although existing process mining techniques can be used to calculate various statistics from event logs for performance analysis, most techniques calculate and display the statistics for each cohort separately. Furthermore, the numerical statistics and simple visualisations may not be intuitive enough to allow users to compare the performance of various cohorts efficiently and effectively. We developed a novel visualisation framework for log-based process performance comparison to address these issues. It enables analysts to quickly identify the performance differences between cohorts. The framework supports the selection of cohorts and a three-dimensional visualisation to compare the cohorts using a variety of performance metrics. The approach has been implemented as a set of plug-ins within the open source process mining framework ProM and has been evaluated using two real-life data sets from the insurance domain to assess the usefulness of such a tool. This paper also derives a set of design principles from our approach which provide guidance for the development of new approaches to process cohort performance comparison.

Real-time video event detection in crowded scenes using MPEG derived features

Investigate multiple instance learning and motion features for event detection.A novel trajectory feature descriptor from the MPEG domain is proposed.A novel multiple instance learning approach using sparse approximation is proposed.Real time performance is achieved. This paper presents an investigation into event detection in crowded scenes, where the event of interest co-occurs with other activities and only binary labels at the clip level are available. The proposed approach incorporates a fast feature descriptor from the MPEG domain, and a novel multiple instance learning (MIL) algorithm using sparse approximation and random sensing. MPEG motion vectors are used to build particle trajectories that represent the motion of objects in uniform video clips, and the MPEG DCT coefficients are used to compute a foreground map to remove background particles. Trajectories are transformed into the Fourier domain, and the Fourier representations are quantized into visual words using the K-Means algorithm. The proposed MIL algorithm models the scene as a linear combination of independent events, where each event is a distribution of visual words. Experimental results show that the proposed approaches achieve promising results for event detection compared to the state-of-the-art.

Unusual Event Detection in Crowded Scenes Using Bag of LBPs in Spatio-Temporal Patches

Modelling events in densely crowded environments remains challenging, due to the diversity of events and the noise in the scene. We propose a novel approach for anomalous event detection in crowded scenes using dynamic textures described by the Local Binary Patterns from Three Orthogonal Planes (LBP-TOP) descriptor. The scene is divided into spatio-temporal patches where LBP-TOP based dynamic textures are extracted. We apply hierarchical Bayesian models to detect the patches containing unusual events. Our method is an unsupervised approach, and it does not rely on object tracking or background subtraction. We show that our approach outperforms existing state of the art algorithms for anomalous event detection in UCSD dataset.

Detecting rare events using Kullback-Leibler divergence

One main challenge in developing a system for visual surveillance event detection is the annotation of target events in the training data. By making use of the assumption that events with security interest are often rare compared to regular behaviours, this paper presents a novel approach by using Kullback-Leibler (KL) divergence for rare event detection in a weakly supervised learning setting, where only clip-level annotation is available. It will be shown that this approach outperforms state-of-the-art methods on a popular real-world dataset, while preserving real time performance.

Discovering work prioritisation patterns from event logs

Business process improvement initiatives typically employ various process analysis techniques, including evidence-based analysis techniques such as process mining, to identify new ways to streamline current business processes. While plenty of process mining techniques have been proposed to extract insights about the way in which activities within processes are conducted, techniques to understand resource behaviour are limited. At the same time, an understanding of resources behaviour is critical to enable intelligent and effective resource management - an important factor which can significantly impact overall process performance. The presence of detailed records kept by todays organisations, including data about who, how, what, and when various activities were carried out by resources, open up the possibility for real behaviours of resources to be studied. This paper proposes an approach to analyse one aspect of resource behaviour: the manner in which a resource prioritises his/her work. The proposed approach has been formalised, implemented, and evaluated using a number of synthetic and real datasets.

An Efficient and Robust System for Multiperson Event Detection in Real-World Indoor Surveillance Scenes

Due to the popularity of security cameras in public places, it is of interest to design an intelligent system that can efficiently detect events automatically. This paper proposes a novel algorithm for multiperson event detection. To ensure greater than real-time performance, features are extracted directly from compressed MPEG video. A novel histogram-based feature descriptor that captures the angles between extracted particle trajectories is proposed, which allows us to capture motion patterns for multiperson events in the video. To alleviate the need for fine-grained annotation, we propose the use of labeled latent Dirichlet allocation, a weakly supervised method that allows the use of coarse temporal annotations, which are much simpler to obtain. This novel system is able to run at ~10 times real time, while preserving state-of-the-art detection performance for multiperson events on a 100-h real-world surveillance data set (TRECVid surveillance event detection).

Activity Modelling in Crowded Environments: A Soft-Decision Approach

Probabilistic topic models have recently been used for activity analysis in video processing, due to their strong capacity to model both local activities and interactions in crowded scenes. In those applications, a video sequence is divided into a collection of uniform non-overlaping video clips, and the high dimensional continuous inputs are quantized into a bag of discrete visual words. The hard division of video clips, and hard assignment of visual words leads to problems when an activity is split over multiple clips, or the most appropriate visual word for quantization is unclear. In this paper, we propose a novel algorithm, which makes use of a soft histogram technique to compensate for the loss of information in the quantization process, and a soft cut technique in the temporal domain to overcome problems caused by separating an activity into two video clips. In the detection process, we also apply a soft decision strategy to detect unusual events. We show that the proposed soft-decision approach outperforms its hard decision counterpart in both local and global activity modelling.

Exposing Impediments to Insurance Claims Processing

(a) Situation faced: Processing injury-compensation claims, such as compulsory third party (CTP) claims, is complex, as it involves negotiations among multiple parties (e.g., claimants, insurers, law firms, health providers). Queensland’s CTP program is regulated by the Motor Accident Insurance Commission (MAIC). The Nominal Defendant, an arm of MAIC, determines liability for claims when the vehicle “at fault” is unregistered or unidentified and manages such claims from injured persons. While the relevant legislation mandates milestones for claims processing, the Nominal Defendant sees significant behavioral and performance variations in CTP claims processing, affecting the costs and durations of claims. The reasons for these variations are poorly understood. (b) Action taken: The BPM initiative took a process-mining approach that focused on the process identification, discovery, and analysis phases of the BPM Lifecycle. We undertook automated process discovery and comparative performance analysis with the aim of identifying where claims processing across cohorts of interest to the Nominal Defendant differed. In parallel, we conducted a context analysis with the aim of identifying the context factors that affect claim duration and cost. The personal injury literature and interviews with representative Nominal Defendant staff informed our selection of data attributes. (c) Results achieved: Process models were developed to facilitate comparative visualization of processes. The Nominal Defendant was particularly interested in differences in the processes for specific cohorts of claims: (i) overall claims, (ii) claims involving unregistered vehicles versus unidentified vehicles, and (iii) direct claims versus legally represented claims. The model facilitated identification of aspects of claims processing where there were significant differences between cohorts. Data mining/feature selection techniques identified a set of process-related context factors affecting claim duration and cost. Models utilizing these context factors were able to distinguish between cases with short and long durations with 68% accuracy and between low-cost and high-cost claims with 83% accuracy. (d) Lessons learned: This multi-faceted process-mining study presented many challenges and opportunities for refining our process-mining methodology and toolset. Data-related challenges arose because of the replacement of claims-management software during the study. Legislative changes, changes to key personnel, and the semi-structured nature of CTP claims-processing introduced issues related to concept drift. Each of these issues affected process discovery, but close collaboration with the stakeholders proved valuable in addressing these issues. Novel visualization techniques were developed to support delivery of insights gained through comparative analysis that will guide process improvement. Consideration of context considerably broadens the scope of process mining and facilitates reasoning about process specifics.