People Tracking

michaelsalerno

Dear Luxonis Team and @jakaskerl ,

I am curious how to load in the following experiment that is listed on the Depthai Experiments Page of github. Below I have left the URL to the website, but I am curious how to run the example for myself. When I install the requirements than run the python main.py file, It loads in the example used on the website. How can I activate my camera while still using the AI model to track people who appear in the cameras point of view? Please let me know if you need me to provide more information.

luxonis/depthai-experimentstree/master/gen2-people-tracker

jakaskerl

Hi @michaelsalerno
Just remove the replay argument: luxonis/depthai-experimentsblob/master/gen2-people-tracker/main.py#L10C1-L10C58

Thanks,
Jaka

michaelsalerno

@jakaskerl ,

Thank you for your help.

From Michael Salerno

michaelsalerno

@jakaskerl

Is there a way to visualize the confirmation percent in the binding boxes? I am just curious as to how I can test the accuracy of the model. Exporting some numbers or figures?

From Michael Salerno

jakaskerl

michaelsalerno confirmation percent

You mean the confidence score that the object detected is correct? I think the yolo should output results (detections) in x, y, width, height, class, confidence so you can use the score from that and display it beside the bbox.

Thanks,
Jaka

michaelsalerno

@jakaskerl ,

Exactly, My only fear with the yolo segmentation is that I am worried it will begin to classify different objects apart from people. I want the same effect and display this has, but I would like the confidence with an x, y, and potential width height. Would I just need to replace or place some lines of code in the main.py virtual studio file? Or is there already an existing experiment with these?

Also is there a way to output the confidence intervals or other detections?

jakaskerl

Hi @michaelsalerno
Yeah, just update to callback to iterate through packet.detections list and display each detection.confidence.

michaelsalerno Also is there a way to output the confidence intervals or other detections?

I'm not sure what you mean.

Thanks,
Jaka

michaelsalerno

@jakaskerl

Would I need to

from packet.detection import …

detection.confidence ()

I am just confused on placement of the code.

Currently I am working with the code and I am trying to take the people tracking experiment and combine it with the people counting. I would also like to include the confidence thresholds as well if possible. The X, Y, and width are not as critical but would still be important. Below is a photo of the current code I am working on. I am having a little bit of trouble activating the counting aspect, but I am able to get the display working.

I am interested to know if I can export the confidence intervals or somehow test the NN to show its accuracy.

jakaskerl

Hi @michaelsalerno
detection.confidence is an element of packet.detections list. Each frame will have many elements if there are multiple detections present.

I suggest you look at: luxonis/depthaiblob/main/depthai_sdk/src/depthai_sdk/classes/packets.py#L381-L445

Thanks,
Jaka