In order to use Accuware Sentinel you need one or more video streams that satisfy these requirements:
- Camera mounting: videos should be recorded from static cameras mounted above head height such that they point on the test area with an angle of about 45°.
- Resolution: the algorithm can process any resolution. To optimize the processing time we suggest to resize the video to the lowest resolution where people can be detected:
- Monitoring a close up view – even a resolution of 320 x 240 px may be sufficient.
- Monitoring a busy scene – at least a resolution of 1280 x 720 px / 720P / HD is required.
- People dimension: the Sentinel algorithm resizes the original video stream to 1366 x 768 pixels before searching for people. In order to be detected, people must have a minimum screen width of at least 30 pixels in the resized video stream. So, even if the resolution of the original video stream is super high, it could be be that people inside the resized video stream are so small that can’t detected (usually those in the background of the scene).
- Frame rate: at least 7 fps (10 fps or more is recommended).
- Field of view (FOV): fields of view up to 90° are supported. Fish-eye lens are not supported!
- Color depth: color streams are recommended for best results.
- Camera type: Sentinel can process video streams from digital and analog cameras. Analog camera streams must go through a DVR to provide remote access.
- Scene lighting: must have good visible lighting conditions. Works fine in outdoors settings unless there are extreme lighting conditions like sun shining directly on the camera. Infrared streams are not supported.
- Accessibility of video streams: video stream must be accessible remotely through a URL. Processing of standalone video files is supported for integration with analog cameras and other video management systems.
- Bandwidth: it depends on resolution of the video and the frame rate. Example:
- a stream 1280 x 720 Px @ 25 FPS requires about 1.5 GB/hr which means a bandwidth of 3.33 Mbps.
Good examples of video inputs