Local Event Detection Research Articles

Temporal analysis for fast motion detection in a crowd

We present a fast motion detection technique in a crowd as an abnormal motion based on optical flow and a motion history image (MHI). Since a camera view is usually not in perpendicular with motion direction, the velocity of motion is not uniform spatially. Instead of object detection directly from an image, we separate an image into several blocks. In this paper, we propose a novel method to analyze a motion using MHI representation, called a shift space and a shift histogram. Together with a velocity histogram, the method can detect fast motion in a crowd, realizing local abnormal event detection. The performance of the proposed method is experimentally illustrated and evaluated.

Measuring the Impact of Engine Oils and Fuels on Low-Speed Pre-Ignition in Downsized Engines

© 2014 SAE International. One of the limits on the maximum fuel efficiency benefit to be gained from turbocharged, downsized gasoline engines is the occurrence of low speed pre-ignition (LSPI). LSPI may lead to high pressures and extreme knock (megaknock or superknock) which can cause severe engine damage. Though the mechanism leading to megaknock is not completely resolved, LSPI is thought to arise from local auto-ignition of areas in the cylinder which are rich in low ignition delay “contaminants” such as engine oil and/or heavy ends of gasoline. These contaminants are introduced to the combustion chamber at various points in the engine cycle (e.g. entering from the top land crevice during blow-down or washed from the cylinder walls during DI wall impingement). This paper describes a method for testing the propensity of different contaminants to cause a local pre-ignition in a gasoline engine. During one cycle, a small amount of contaminant is injected into one cylinder of a 4 cylinder engine. The spark is suppressed during this or the following cycle to allow detection of local pre-ignition events after spark timing. If the contaminant is injected on the cycle before, it is the cycle following the injection that has the missed spark. By detecting auto-ignition events before and after spark timing, it is possible to compare contaminants over a broad range of ignition tendencies. Sensitivities of pre-ignition tendencies to intake pressures and temperatures, the amount of contaminant introduced, and fuelling ratios are discussed. Additionally, the importance of contaminant stratification is shown, and pre-ignition is demonstrated to result from the introduction of contaminant as early as the beginning of blow-down of the preceding cycle.

Text-Based Twitter User Geolocation Prediction

Geographical location is vital to geospatial applications like local search and event detection. In this paper, we investigate and improve on the task of text-based geolocation prediction of Twitter users. Previous studies on this topic have typically assumed that geographical references (e.g., gazetteer terms, dialectal words) in a text are indicative of its author’s location. However, these references are often buried in informal, ungrammatical, and multilingual data, and are therefore non-trivial to identify and exploit. We present an integrated geolocation prediction framework and investigate what factors impact on prediction accuracy. First, we evaluate a range of feature selection methods to obtain “location indicative words”. We then evaluate the impact of non-geotagged tweets, language, and user-declared metadata on geolocation prediction. In addition, we evaluate the impact of temporal variance on model generalisation, and discuss how users differ in terms of their geolocatability. We achieve state-of-the-art results for the text-based Twitter user geolocation task, and also provide the most extensive exploration of the task to date. Our findings provide valuable insights into the design of robust, practical text-based geolocation prediction systems.

Text-based twitter user geolocation prediction

Geographical location is vital to geospatial applications like local search and event detection. In this paper, we investigate and improve on the task of text-based geolocation prediction of Twitte...

Continuous monitoring of global events in sensor networks

Event detection is an essential application in Wireless Sensor Networks (WSNs), especially for the monitoring of physical world. While most previous research focuses on the detection of local events, in this paper we propose novel algorithms to energy efficiently detect events of large and global scale. We divide a global event into regional events, so the detection algorithm can be executed distributively inside the network. Our approach also takes advantage of the temporal correlations of sensing data to gain energy efficiency. It uses a bound-suppression mechanism to set bounds and suppresses 'silent' regional events, cutting down the transmissions. Simulation results show that the proposed c-GEDA (Continuous Global Event Detection Algorithm) is efficient to reduce the cost of transmissions, and it gains more than 50% of cost reduction compared with the previous detection approaches.

An automated local and regional seismic event detection and location system using waveform correlation

AbstractWe report on the development of an automated Local Waveform Correlation Event Detection System (LWCEDS) and its application to the New Mexico Tech Seismic Network. LWCEDS is an adaptation of a global system, WCEDS, a matched filtering algorithm for global Comprehensive Test Ban Treaty (CTBT) monitoring applications developed at Sandia National Laboratories and New Mexico Tech. Although the current CTBT monitoring system is based on teleseismic phase detection, effort is being placed on research to highlight specific areas of the globe for which local and regional seismic networks could be employed. An automated waveform correlation regional location system could also serve as a rapid alert and automated location system by providing magnitude and hypocenter information within a few minutes of the occurrence of a hazardous earthquake.In the LWCEDS algorithm, processed waveforms are correlated with theoretical travel-time envelopes, and a grid search is performed to identify the space-time solutions that yield the highest correlations. High correlation indicates that an event has occurred and that a good approximation to the correct origin time and hypocenter has been determined; explicit phase identification is not required. To avoid the large computational expense of calculating a complete correlation for each grid point, we use a laterally homogeneous velocity model and reformulate the problem into a single matrix multiplication and matrix assessment for each time step. LWCEDS has been successfully tested on a suite of local and regional seismic events selected to span the range of expected event quality. Preliminary results from our sparse network show typical epicentral errors of less than 3 km for local events and, with notable exceptions, to within 10 to 20 km for regional events. Similar results were obtained during an on-line experiment conducted to generate daily bulletins during the time period from 3 December 1996 through 7 January 1997. Data from 156 triggers were processed, including 33 teleseisms, 102 regional events (includes explosions), and 21 local earthquakes. Results from this expanded test set are encouraging but reveal the need for a method to mask or flag various electronic and telemetry spikes.

Trends in Severe Local Storm Watch Verification at the National Severe Storms Forecast Center

Abstract Trends of tornado and severe thunderstorm watch verification for the period 1967–1990 are presented. Over the past 10 years the annual number of reported severe thunderstorm events has increased substantially. In comparison, the number of tornado events reported has remained relatively constant from year to year. During the period 1967 to 1990, the percentage of watches verifying has increased from 45% to 85%, while the probability of detection (POD) of severe local storm events has increased from about 0.35 to 0.50. Recent results show that the number of severe thunderstorm watches has steadily increased since 1985, the size of watches has slowly decreased, and lead time has slightly decreased. Yearly fluctuations in tornado watch verification statistics appear to be best related to the number of outbreak tornadoes. Data suggest the Severe Local Storm Unit (SELS) performance in forecasting outbreak episodes and strong/violent tornadoes is improving at a faster rate than in forecasting isolated t...

Local biotransformation of des-Tyr 1-γ-endorphin in brain studied by a push-pull technique and HPLC analysis

The biotransformation of [ 3H]des-Tyr 1-γ-endorphin locally in rat brain was studied by perfusion of the peptide via a push-pull cannula implanted in the nucleus caudatus. Metabolites were analyzed by high-pressure liquid chromatography. Free [ 3H]phenylalanine and a [ 3H]peptide with chromatrographic properties of β-endorphin 2–13 were formed during push-pull perfusion. The method presented enables the detection of local proteolytic events in the brain and provides a novel tool for studies on neuropeptide biotransformation.

Real-time detection and location of local seismic events in central California

Abstract A computer-based system dedicated full time to automatic detection and location of local seismic events in central California has been developed. The system monitors 108 short-period vertical-component stations from the U.S. Geological Survey central California and Oroville seismic networks. Locations and magnitudes, when determined, are printed out along with first arrival times, within 2 to 5 min after an event occurs. Wave onsets must be clear and impulsive for best results. For this reason, regional events and teleseisms are usually rejected. The best results have been obtained for the relatively dense, 16-station Oroville network. For the month of October 1975, 107 (91 per cent) of the 118 events timed by hand were also timed and located by the real-time system. An additional eight events (7 per cent) were detected in real-time but were not successfully located. Of the 107 events for which both on-line and hand-timed locations are available, 92 per cent of the on-line locations are within 2 km of the epicenters determined by hand-timing. During October 1975 the real-time system monitored 91 of the 150 stations of the central California network. Of the 260 events located by hand-timing, 225 (86 per cent) were detected by the real-time system. Magnitudes of detected events range from 0.8 to 2.9. Approximately 95 per cent of the events of magnitude 1¼ and greater detected and located by hand-timing methods were also detected by the real-time system. Differences between hypocentral locations based on hand-timed and computer-timed arrivals may vary from 0.1 to 5 min of latitude or longitude.