Metrici Thermal Analyzer engine

Distributed thermal temperature-measurement engine

Features

Architecture: Distributed detection and recognition, free-flow or triggered
Number of recognized faces in each image: Unlimited (the width of each face should be more than 4% of the width of the image)
Individual temperature measurement for each face: Yes
Working mode: Single or dual view (visual + thermal, or thermal + thermal)
Thermal sensor selection: Yes (left, right or both)
Recognition on thermal image: Yes (it should be used because it cannot be fooled by a printed face)
Recognition on visible image: Yes (the parallax error should be corrected at installation time)
Face bounding box reduction: Yes (up to 90%)
Number of thermal algorithms: 4 (average temperature, average of the hottest 10% pixels, average of the hottest 1% pixels, hottest pixel)
Number of detection zones: Unlimited
Alarm temperatures range: Yes (min and max can be defined)
Signaling device: Yes (the ON URL, the OFF URL, and the delay can be defined individually)
Companion image: Yes (from another camera or from the same camera using the JPEG image driver)
Temperature calibration: Yes (static or by using a black body)
Type of objects recognized: Human heads and human bodies (the neural network can be trained to recognize other types of objects, like industrial parts or animals)

System requirements

OS: Ubuntu 24.04 LTS MATE, Linux CentOS 7 (64-bit)
CPU: Intel Xeon; AMD Ryzen, Threadripper or Epyc
GPU: Intel, AMD or Nvidia; mandatory for more than 2 cameras/server
System memory: 512 MB for each connected camera, but no less than 4 GB/server
GPU memory: At least 256 MB available for each connected camera
Storage: Up to 512 KB for each thermal event stored in the buffer, depending on the camera resolution and image compression
Cameras/server: Up to 256 connected cameras on each server; unlimited number of servers
Camera type: IP; MxPEG video stream; HTTP transport protocol
Supported camera producers: Mobotix