I'm using the BLIP2 board and the AD Vision Sensor Controller GUI. For the indoor occupancy, are the red numbers in the upper right hand corner the estimated range number of occupants? Is this based on blob detection or another type of algorithm?
Also, when using the AD Vision Sensor GUI, the bounding boxes around people will not show up, even when I ensure to select them to be displayed in the GUI. What could be causing this problem? Are there any tuning parameters you recommend changing?
Yes the red numbers are indeed the estimated occupancy range in the current field of view. This is influenced by motion and moving chairs and will give a more accurate count once everyone settles down. For more accurate results this has to be fixed in an overhead configuration and the width and height parameter has to be tuned. Please see this thread for more details -
BLIP board Evaluation
Regarding the bounding box display the parameter you have to enable is Enable graphics. Are you having an issue in getting all Bounding boxes to show up or some people not getting detected. Can you let me know the configuration it has been setup in?. If it is setup too high the size of blobs to be detected will have to be reduced. I can give you the exact parameters to use based on the configuration of setup.
Thanks, that helps a lot. So would the red numbers match the number of blobs detected in the VOSBlobsInfo file?
Making sure Enable Graphics is 1 did solve my bounding boxes problem. Thanks! Also, since my set up is a top down configuration about 8.5 ft high, would you recommend increasing the height and width parameters of a person for better results? What else can I tune to improve performance or calibrate the system? I noticed that it does pick up chairs and other items as people.
The red numbers or the occupancy estimate are arrived at using a post processing on the detected blobs so may not exactly match to the contents of the VOSBlobsInfo file.
The width and height parameters correspond to the height and width in pixels that a detected person would appear in the view so if it appears that a person is detected as two in the field of view, please try increasing this. I feel the current default parameters will be fine for the lens and setup.
It may indeed pickup chair and other motion but will take some time to settle down and eventually (after the coasting period) give a more appropriate estimate.
Thanks. So if I am evaluating the algorithms performance for occupant counting, it would be better to use the red numbers then, correct? Is the occupancy estimate recorded anywhere else, such as in a log file?
Unfortunately it is not logged to a file, but we do send it over SPI using the WSN connection if that would work. We are working sending this over other interfaces like UART to enable logging in the next release but currently I think it is only logged onto the video frame.