Take pictures of a detection and store these pictures (detections)

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Hello,
I would like to control an actuator system and take parallel pictures when detecting a model that has already been specially trained. How can I implement this.
For example, the following application would be possible: As soon as a water bottle is visible in the image, take a photo, save this photo and output a signal via an LED. This storing of the photo and the output of the LED signal should take place simultaneously. How can I implement this?

Nvidia Jetson TX2.

2

Hi markusnvidia,

That project sounds interesting, you will require an inference engine that detects the bottles, then after each detection a signal should be sent to take the snapshot and blink the led.

We have an open source project called GstInference which handles object detection (TinyYOLOV2 and TinyYoloV3). We created a GStreamer plugin that activates signals after each object detected.
This can be modified to exclusively detect bottles.

This is the link to the project’s page: https://developer.ridgerun.com/wiki/index.php?title=GstInference/Introduction

This is an possible implementation for your case.

Greivin F.

3

Hi,

You can also refer below link in case it helps:

Thanks

4

Hello,

is there an oppurtunity to take pictures via Python with gstCamera?

5

Hi,
Please refer below link:
https://github.com/dusty-nv/jetson-inference/blob/master/python/examples/imagenet-camera.py#L51

Thanks

6

Thanks. But how can I store this image on my drive within the python code?

7

Hi,

Please refer below link:

Thanks

8

Thanks, it works.
Now I have the problem that only one picture is saved at a time. Every further image overwrites the already existing one, because I have defined a fixed name for the image. Can I keep the name of the image variable. So if the name already exists, that another file name is chosen?

Thanks.

9

Yes, that should work. You can even add timestamp to output file name to avoid overwrite issue.

Thanks

10

How can I add timestamp to output file?

11

Please refer below links to get some idea:
https://stackoverflow.com/questions/10607688/how-to-create-a-file-name-with-the-current-date-time-in-python
https://stackoverflow.com/questions/5218261/python-how-to-rename-a-text-file-with-datetime/5218328

Thanks

12

Thanks. It works. I´ve another question.
Is there an oppurtunity to save the coordinates of the detected objects (maybe in a .txt format)?

13

Hi,

Please refer to below detectNet links:

https://rawgit.com/dusty-nv/jetson-inference/python/docs/html/python/jetson.inference.html#detectNet

Thanks

14

Thank you. I have trained my own models with DIGITS.
I cannot use them in this code:

net = jetson.inference.detectNet("my-own-model", threshold=0.5)

This is only for pretrained models. How can I use my own model in my “own code”?

15

Hi,

Please refer to below link:
https://github.com/dusty-nv/jetson-inference/blob/f6171be1c25f239180666bb0e3193831d9a5e923/docs/imagenet-custom.md

Thanks

17

How can I use my own model in this line?
net = jetson.inference.detectNet(opt.network, sys.argv, opt.threshold)

These are the files, I have in my own created model directory:Screenshot from 2020-04-07 11-46-06

18

I want to take a photo when I detect an object, how can I do it?

I use docker and my usb camera (/dev/video0)

19

How do i combine my coding for image detection with led signal?