Description:
This paper presents iNNE (isolation using Nearest Neighbour Ensemble), an efficient nearest neighbour-based anomaly detection method by isolation. Inne runs significantly faster than existing nearest neighbour-based methods such as Local Outlier Factor, especially in data sets having thousands of dimensions or millions of instances. This is because the proposed method has linear time complexity and constant space complexity. Compared with the existing tree-based isolation method iForest, the proposed isolation method overcomes three weaknesses of iForest that we have identified, i.e., Its inability to detect local anomalies, anomalies with a low number of relevant attributes, and anomalies that are surrounded by normal instances.
Description:
Nearest neighbour search is a core process in many data mining algorithms. Finding reliable closest matches of a query in a high dimensional space is still a challenging task. This is because the effectiveness of many dissimilarity measures, that are based on a geometric model, such as lp-norm, decreases as the number of dimensions increases. In this paper, we examine how the data distribution can be exploited to measure dissimilarity between two instances and propose a new data dependent dissimilarity measure called 'mp-dissimilarity'. Rather than relying on geometric distance, it measures the dissimilarity between two instances in each dimension as a probability mass in a region that encloses the two instances. It deems the two instances in a sparse region to be more similar than two instances in a dense region, though these two pairs of instances have the same geometric distance. Our empirical results show that the proposed dissimilarity measure indeed provides a reliable nearest neighbour search in high dimensional spaces, particularly in sparse data. Mp-dissimilarity produced better task specific performance than lp-norm and cosine distance in classification and information retrieval tasks.
Description:
Abstract— Corner detectors have many applications in computer vision and image identification and retrieval. Contour-based corner detectors directly or indirectly estimate a significance measure (e.g., curvature) on the points of a planar curve, and select the curvature extrema points as corners. While an extensive number of contour-based corner detectors have been proposed over the last four decades, there is no comparative study of recently proposed detectors. This paper is an attempt to fill this gap. The general framework of contour-based corner detection is presented, and two major issues – curve smoothing and curvature estimation, which have major impacts on the corner detection performance, are discussed. A number of promising detectors are compared using both automatic and manual evaluation systems on two large datasets. It is observed that while the detectors using indirect curvature estimation techniques are more robust, the detectors using direct curvature estimation techniques are faster.
Description:
Haptic perception is mediated by tactile (skin-surface stimulation) and kinesthetic (perception of movement indicating limb position) inputs. In a recognition experiment, participants were asked to match stimuli of different curvatures to those in a visual choice set. They used tactile-only information, kinesthesis-only information, and a combination of both. Movements were either self-generated or passively-guided. The experiment showed that judgements based on self-generated kinesthetic information were more accurate than those based on tactile-only information.
Description:
Objectives: To evaluate the current evidence regarding the quantification of cricket fast-bowling skill. Methods: Studies that assessed fast-bowling skill (bowling speed and accuracy) were identified from searches in SPORTDiscus (EBSCO) in June 2017. The reference lists of identified papers were also examined for relevant investigations. Results: A total of 16 papers matched the inclusion criteria, and discrepancies in assessment procedures were evident. Differences in test environment, pitch, and cricket ball characteristics; the warm-up prior to test; test familiarization procedures; permitted run-up lengths; bowling spell length; delivery sequence; test instructions; collection of bowling speed data; and collection and reportage of bowling accuracy data were apparent throughout the literature. The reliability and sensitivity of fast-bowling skill measures have rarely been reported across the literature. Only 1 study has attempted to assess the construct validity of its skill measures. Conclusions: There are several discrepancies in how fast-bowling skill has been assessed and subsequently quantified in the literature to date. This is a problem, because comparisons between studies are often difficult. Therefore, a strong rationale exists for the creation of match-specific standardized fast-bowling assessments that offer greater ecological validity while maintaining acceptable reliability and sensitivity of the skill measures. If prospective research can act on the proposed recommendations from this review, then coaches will be able to make more informed decisions surrounding player selection, talent identification, return to skill following injury, and the efficacy of short- and long-term training interventions for fast bowlers.
Description:
In passive localization, sensors try to locate an event without any knowledge of event's emitted power. So, this is a more challenging problem compared to active localization. Existing passive localization schemes use expensive and noise-vulnerable range-based techniques. In this paper, we propose, to the best of our knowledge for the first time, a cost-effective range-free passive localization scheme exploiting hybrid sensor network model where mobile sensors are deployed on demand once an event is sensed by a static sensor. Efficient use of mobile sensors leads to two concomitant optimization problems: (1) positioning the mobile sensors so that the expected possible event location area is minimized; and (2) minimizing their overall traversed distance. To solve the first problem, we have developed a novel arc-coding based range-free localization technique that can accurately define the area of possible event location from the feedback of arbitrarily placed sensors without relying on expensive hardware to estimate range of signals. We have achieved significantly high localization accuracy with a low number of mobile sensors even after considering significant environmental noise. To solve the second problem, three alternative deployment strategies for the mobile sensors were simulated to recommend the best.
Description:
In many application scenarios, wireless sensors are deployed deterministically throughout a wide area to detect and report specific events or monitor environmental parameters. To cover a large area with minimal number of sensors, it is important to determine sensing radius of the operating sensors. Since the emitted energy of a random event is neither predictable nor fixed, accurate sensing radius modelling is a challenging problem. To the best of our knowledge, no work has considered how the event intensity factor reduces probability of event detection while assuming a sensing radius despite its high significance in important areas such as coverage, detection, localization, etc. In this paper, we have proposed a novel stochastic model of the maximum sensing radius to guarantee a user-defined event detection probability from the pdf of average event intensity and the quality of sensors. Comprehensive theoretical and numerical analyses are presented to evaluate the event detection performance of this model against different relevant parameters and these are also verified by simulation. Provision for event location trajectory computation is analysed for high-intensity events.
Description:
The problem of recognizing human actions is characterized by complex dynamics and strong variations in their executions. Despite this inconvenience, space-time correlations provide valuable clues for their discrimination. Therefore, space-time correlators like emph{Maximum Average Correlation Height} (MACH) filters have successfully been used for action recognition with encouraging results. However, their utility is challenged due to number of factors: (i) these filters are trained only for one class at a time and separate filters are required for each class increasing computational overhead, (ii) these filters simply take average of similar action instances and behave no better than average filters and (iii) misaligned action datasets create problems for these filters as they are not shift-invariant. In this paper, we address these issues by posing action recognition as a multi-class discrimination problem and propose a emph{single} 3D frequency domain filter, named Action ST-DCCF for multiple action classes that mitigates inherent discrepancies of correlation filters. It presents a different interpretation of correlation filters as a method of applying spatio-temporal transformation to the data rather than simply minimizing correlation energy across all possible shifts. Experiments on a variety of action datasets are performed to evaluate our approach. Experimental results are comparable to the existing action recognition approaches.
Description:
The problem of recognizing human actions is characterized by complex dynamics and strong variations in their executions. Despite this inconvenience, space-time correlations provide valuable clues for their discrimination. Therefore, space-time correlators like \emph{Maximum Average Correlation Height} (MACH) filters have successfully been used for action recognition with encouraging results. However, their utility is challenged due to number of factors: (i) these filters are trained only for one class at a time and separate filters are required for each class increasing computational overhead, (ii) these filters simply take average of similar action instances and behave no better than average filters and (iii) misaligned action datasets create problems for these filters as they are not shift-invariant. In this paper, we address these issues by posing action recognition as a multi-class discrimination problem and propose a \emph{single} 3D frequency domain filter, named Action ST-DCCF for multiple action classes that mitigates inherent discrepancies of correlation filters. It presents a different interpretation of correlation filters as a method of applying spatio-temporal transformation to the data rather than simply minimizing correlation energy across all possible shifts. Experiments on a variety of action datasets are performed to evaluate our approach. Experimental results are comparable to the existing action recognition approaches.
Description:
Wireless sensor networks promise to guarantee accurate, fault tolerant and timely detection of events in large scale sensor fields. To achieve this the notion of k-coverage is widely employed in WSNs where significant redundancy is introduced in deployment as an event is expected to be sensed by at least k sensors in the neighborhood. As sensor density increases significantly with k, it is imperative to find the optimal k for the underlying event detection system. In this work, we consider the detection probability, fault tolerance and latency as the Quality of Service (QoS) metrics of an event detection system employing k-coverage and present a probabilistic model to guarantee given QoS support with the minimum degree of coverage taking into account the noise related measurement error, communication interference and sensor fault probability. This work eventually resolves the problem of over or under deployment of sensors, increases scalability and provides a well defined mechanism to tune the degree of coverage according to performance needs.
Description:
Wireless sensor networks promise to guarantee accurate, fault tolerant and timely detection of events in large scale sensor fields. To achieve this the notion of k-coverage is widely employed in WSNs where significant redundancy is introduced in deployment as an event is expected to be sensed by at least k sensors in the neighborhood. As sensor density increases significantly with k, it is imperative to find the optimal k for the underlying event detection system. In this work, we consider the detection probability, fault tolerance and latency as the Quality of Service (QoS) metrics of an event detection system employing k-coverage and present a probabilistic model to guarantee given QoS support with the minimum degree of coverage taking into account the noise related measurement error, communication interference and sensor fault probability. This work eventually resolves the problem of over or under deployment of sensors, increases scalability and provides a well defined mechanism to tune the degree of coverage according to performance needs.