"Intelligent Monitoring and Automatic Detection of Threats" accomplished a phase of prototype testing
WP1 (Work Package 1) algorithms and tools for audio and video processing and event detection were evaluated both in real conditions (documented in deliverable D1.3 “Document reporting on acquired results of pilot trial”) and in laboratory conditions, based on benchmark recordings (subject of the deliverable D1.4 "Multimedia database documentation with analysis of recommended algorithms"). Algorithms were developed by WP1 Partners, and their goal is to support intelligent detection of potentially dangerous events. These tools exist as several prototypes and modules: 1) a framework developed by GUT, comprising Node Stations and Central Station and extended with other Partners modules; 2) standalone algorithms run within developer environments.
Deliverable D1.3 contains a summary of pilot trials of tools, modules, and prototypes of WP1. Each subsection focuses on one particular test installation (test-bed location), describing purpose of the trial, algorithms tested in those conditions, procedures of data registration and data analysis, and summarising results. Following tests were conducted, and their outcomes are documented: 1) related to persons: counting persons entering and leaving a building for detection of people remaining inside after working hours, detection of an unattended luggage; 2) related to parking lot management: counting vehicles entering or leaving a parking lot, detection of the parking vehicle and parking place identification (i.e. when and where a vehicle has parked), detecting vehicle stopping outside of any parking place (i.e. on the road), 3) detecting traffic events related to prohibited and dangerous lane changing; 4) related to audible manifestation of potentially dangerous evens: scream, gunshot, explosion, breaking glass events localization by sound analysis; 5) object classification into classes: human, vehicle, and other, used for traffic events such as detection of person on the road or a vehicle hitting road barrier, and used for abandoned luggage detection; 6) examination of video streaming quality and bit-rate impact on effectiveness of operator-performed tasks.
Documentation of developed algorithms can be found in deliverable D1.4, which provides also recommendations and guidelines for their utilization, and suggesting optimal values for algorithm parameters.
Developed modules can be arranged into several classes:
1. Algorithms already running in real-time on the basis of WP1 Node Station framework:
- Threat detection, verified in several test conditions (indoor, cloudy and sunny outdoor), for entering a protected area, e.g. person on a railroad tracks,
- Dangerous tool tracking with varying object sizes and orientations in source and destination cameras, for validation in varying conditions,
- Parking events detection,
- Dangerous tool tracking, ditto, e.g. person going towards airport apron,
- Traffic events detection, e.g. dangerous lane changing,
- Object classification, useful in e.g. luggage detection, person on the road, car hitting roadside
- Audio event detection with precise localization of the source, and possibility of multiple concurrent sources detection (gunshot, screaming, explosion, breaking glass)
- 2. Algorithms developed as stand-alone software, running close to real-time, but operating on video files or audio recordings
- Threat detection by means ofstereo cameras, its features allows for object measurements, and 3D localisation, employed in dangerous tool tracking
- Dangerous tool detection, e.g. knife, gun
- Sound type classification module