Grafana dashboards - AI help page

These dashboards display statistics gathered from your analytical pipelines, either within current day or week depending on your selection, but otherwise they work almost the same way.

Every dashboard has two panels with a graph containing specific metrics and a table that displays minimum, maximum and average of those metrics during the selected time frame. Displayed metrics are updated almost in real time. Preset refresh rate of every dashboard is 30 seconds. Every workday has a highlighted time between 6:00 AM and 6:00 PM with a brighter gray background. In weekly statistics weeks start with Sunday and end with Saturday.

Detected objects (DO) - Day / Week

These dashboards display the number of detected objects (DO) on your streams and its anomaly compared to the last three weeks.

In the top left corner it is possible to select pipeline, stream and for this selection show statistics for one or more classes.

First (top) panel contains info about how the number of DOs of each selected class change over a set time frame. It is modified with moving average to help it ignore false positives and then the final value is rounded to integer.

Second (bottom) panel displays an anomaly of selected DO classes, which is represented as a rounded difference between values displayed in the top panel and median of that metric on the same day in each of the 3 previous weeks. If the anomaly is positive, it means that there are more DOs than it is common for that specific time. If it is negative, there are fewer DOs than usual and if the anomaly is equal to zero, the usual number of objects is detected. Anomaly is also generated for future time, but that is irrelevant and should be ignored.

Line Crossing (LC) - Day / Week

These dashboards display the number of objects located inside an area specified by two line crossings (LC) of which one serves as an entrance and the second one as an exit. Also the anomaly of this statistic compared to the last three weeks is displayed. 

In the top left corner it is possible to select one of the available pipelines and for both entrance and exit of the area it is possible to choose a separate source and LC. This option enables entrance and exit to be monitored by different cameras. There can always be only one LC for entrance and one LC for exit of each area. (If the same LC is selected for both entrance and exit, both graphs are going to show a flat line with a value of zero.) Note, that there is no option to specify a class of detected objects here, this can be setup in the analytics settings tab in the pipeline editor (3.2.1.).

First (top) panel contains information about how the number of objects inside the area changes over a set time frame. This number is calculated as the difference between the number of objects that crossed the entrance line and the number of objects that crossed the exit line. If the value is positive, it means that there are objects inside the said area. If the value is zero, there should not be any objects inside the area. If the value is negative, there is a high chance that the entrance and exit LCs don't correspond, or there is another way in or out of the area.

Second (bottom) panel displays an anomaly of the metric from the top panel, which is represented as a rounded difference between values displayed in the top panel and average of that metric on the same day in each of the 3 previous weeks. If the anomaly is positive, it means that there are more objects in this area than it is common for that specific time. If it is negative, there are fewer of these objects than usual and if the anomaly is equal to zero, the usual number of objects is detected inside this area. Anomaly is also generated for future time, but that is irrelevant and should be ignored.

Region of interest (ROI) - Day / Week

These dashboards display the number of objects located inside the selected region of interest (ROI) and also the anomaly of this statistic compared to the last three weeks. Note, that there is no option to specify a class of detected objects here, this can be setup in the analytics settings tab in the pipeline editor (3.2.1.).

In the top left corner it is possible to select one of the available pipelines and also the ROI for which the statistics will be displayed.

Because of the fact that object detection inside a ROI has many different use cases, two types of dashboards exist - max and sum, both in Day / Week versions. Queries for both max and sum versions use Aggregation operators. If only one camera is used to monitor said ROI max and sum dashboards display the exact same statistics.

ROI - Max

In this dashboard, the total number of objects in the selected ROI is displayed. It can also be used when multiple cameras are monitoring one region. In this case, the highest number of detected objects is displayed for every point in time. Every ROI included in this calculation must have the exact same name as the others!

This could be useful when multiple cameras are monitoring the same room from different angles. Certain objects could be obscured for some cameras, so this redundancy minimizes the risk of the statistic being inaccurate.

ROI - Sum

Second type of ROI dashboards is useful when different cameras are monitoring different parts of the same ROI, or when there is more than one ROI and all of them have to be considered in one statistics. In these cases the displayed value is the sum of all the objects detected on all of the used cameras. Every ROI included in this calculation must have the exact same name as the others!

One use case of this is when the ROI is too large to be monitored by just one camera and different cameras have to be monitoring different parts of it. Second example for this use case would be when it is necessary to monitor many similar places like surroundings of every single ATM that some bank owns, different offices, or warehouses and it is useful to know how many objects are being detected at all of these ROIs combined.