PURE

Real-time surveillance video quality detection and restoration for video analysis under difficult environment conditions

Automatic surveillance video analysis is significantly limited by difficult environmental conditions. Changing lighting (day/night, seasons), difficult weather (snow, rain, fog), unstable camera attachment, camera dust and dirt pollution and transmission limitations like low bit rate encoding causes severe degradation of the video quality. This causes non- or bad functioning automatic video surveillance systems and suboptimal conditions for manual inspection.

The goal of the project is to develop novel video quality detection and real-time restoration technologies as a generic and modular pre-processing step for any automatic and manual surveillance video inspection. The restored video quality - also from sources exposed to  difficult environmental conditions - will enable real exploitation of existing and emerging video surveillance infrastructure. The major scientific challenges are to develop restoration and quality detection solutions suited for bad quality outdoor videos and to overcome the tremendous challenge of doing restoration in an online real-time environment.

Results
To detect image sensor and transmission noise in typical surveillance video scenarios (e.g. monitoring highways) we developed a so called "noise level detector". The detector provides reliable estimates of the noise magnitude and furthermore, is also able to estimate the magnitudes of snowfall properly.
For restoration of image sensor and transmission noise in surveillance video we developed the PURE noise reduction filter which provides superior quality results compared to alternative state of the art methods. The filter can also be successfully applied to snowfall suppression.
For restoration of camera shake caused by trucks or heavy wind we built an "image stabiliser" which is able to provide real-time shake free surveillance video.

PURE

Real-time surveillance video quality detection and restoration for video analysis under difficult environment conditions

Automatic surveillance video analysis is significantly limited by difficult environmental conditions. Changing lighting (day/night, seasons), difficult weather (snow, rain, fog), unstable camera attachment, camera dust and dirt pollution and transmission limitations like low bit rate encoding causes severe degradation of the video quality. This causes non- or bad functioning automatic video surveillance systems and suboptimal conditions for manual inspection.

The goal of the project is to develop novel video quality detection and real-time restoration technologies as a generic and modular pre-processing step for any automatic and manual surveillance video inspection. The restored video quality - also from sources exposed to  difficult environmental conditions - will enable real exploitation of existing and emerging video surveillance infrastructure. The major scientific challenges are to develop restoration and quality detection solutions suited for bad quality outdoor videos and to overcome the tremendous challenge of doing restoration in an online real-time environment.

Results
To detect image sensor and transmission noise in typical surveillance video scenarios (e.g. monitoring highways) we developed a so called "noise level detector". The detector provides reliable estimates of the noise magnitude and furthermore, is also able to estimate the magnitudes of snowfall properly.
For restoration of image sensor and transmission noise in surveillance video we developed the PURE noise reduction filter which provides superior quality results compared to alternative state of the art methods. The filter can also be successfully applied to snowfall suppression.
For restoration of camera shake caused by trucks or heavy wind we built an "image stabiliser" which is able to provide real-time shake free surveillance video.

Noise Level Detection                                               Noise Restoration

Snowfall Restoration                                                 Image Stabilization

Results
To detect image sensor and transmission noise in typical surveillance video scenarios (e.g. monitoring highways) we developed a so called "noise level detector". The detector provides reliable estimates of the noise magnitude in real-time and furthermore is also able to estimate the magnitudes of snowfall properly.
For restoration of image sensor and transmission noise in surveillance video we developed the PURE noise reduction filter which provides superior quality results compared to alternative state of the art methods. The filter can also be successfully applied to snowfall suppression.
For restoration of camera shake caused by trucks or heavy wind we built an "image stabiliser" which is able to provide real-time shake free surveillance video.

Automatic surveillance video analysis is significantly limited by difficult environmental conditions. Changing lighting (day/night, seasons), difficult weather (snow, rain, fog), unstable camera attachment, camera dust and dirt pollution and transmission limitations like low bit rate encoding causes severe degradation of the video quality. This causes non- or bad functioning automatic video surveillance systems and suboptimal conditions for manual inspection.

The goal of the project is to develop novel video quality detection and real-time restoration technologies as a generic and modular pre-processing step for any automatic and manual surveillance video inspection. The restored video quality ? also from sources exposed to  difficult environmental conditions - will enable real exploitation of existing and emerging video surveillance infrastructure. The major scientific challenges are to develop restoration and quality detection solutions suited for bad quality outdoor videos and to overcome the tremendous challenge of doing restoration in an online real-time environment.