The FLIR Lepton is a low-resolution LongWave InfraRed (LWIR) camera. The purpose of this article is to explain the main challenges presented by the use of this IR camera in a typical image processing task like automated people counting. Other imaging tasks may present similar or different challenges. RSIP Vision can help you address them in the most professional and effective way.
 

Steps in people counting

People counting involves the following phases:

  • Blob detection – Forming an object out of connected pixels
  • Blob analysis – Classifying blobs as people
  • People tracking – Forming a progress path, within the scene, for each person
  • People counting – The total number of persons inside the scene after each frame update

When using the FLIR Lepton camera the following challenges exist:

1. Quality – Low quality, or noisy image, translates to non-uniform blobs. Their contained pixels intensity may have a large range. Each pixel intensity may be changed dynamically by the camera (from frame to frame), even if the blob is static. The quality may challenge the blob detection. Here is a video example:
 

 

2. Low resolution – Small blobs are difficult to detect and to separate between close ones. Blob detection and analysis is challenged. Here is a video example:

 

 

3. Angle of capturing and people profile – should maximize the people cross section while maintain as larger as possible inner space. In the case of people are bending down or sitting, their coverage profile is altered. This property is challenging the analysis and the tracking phases. Low resolution and noisy image are contributing to higher difficulty. Here is a video example:

 

 

4. Environment status – Emitting (hot) objects may create blobs that will be connected to people blobs. It challenging the analysis phase. Here are two video examples:

 

 

5. Partial figures – When a person is occluded, partially, by a “cold” object, the analysis phase is challenged to handle partial figures. Here are two video examples:

 

 

6. Non-uniform IR signature – People may be constructed of blobs with various gray level contained. For example, clothes will change the IR signature. The actual signature, in this case, is depended of the cloth coverage and the material it made from. The analysis phase is challenged. Here are two video examples:

 

 

7. Low frame rate – In the case of fast object (or very close people, running or fast walkers), the tracking algorithm should be smart to eliminate multiple counting when the object (person) successive blob is detected with larger gap in related to the previous blob location.

8. Mechanical shutter – On the Lepton model with mechanical shutter, additional delay may exist every few frames. The mechanical shutter is closed, for a short duration, to enable internal re-adjustment.

People counting is only one of the near infinite quantity of tasks of image processing that are possible in the real world. RSIP Vision has the experience and skills to advise you for this purpose. We also master state-of-the-art advanced algorithms to help you in all your computer vision projects. Consult with us now.

Share The Story