Bike project with Jetson Nano

Hello again,

Thank you first for answering and your time. I’m still working on the bike counting project, but I run into several problems. At the moment I use a Jetson Nano with USBCam, and after installing it, everything works ok a couple of times. If I turn off and on a couple of times I have the following error and I don’t know how to solve it.

Start recording
already started
First request on console stream
Something went wrong: connect ECONNREFUSED 172.18.0.3:8070
Too many retries, YOLO took more than 3 min to start, likely an error
0
Something went wrong: connect ECONNREFUSED 172.18.0.3:8070
Too many retries, YOLO took more than 3 min to start, likely an error
0
New client, close previous stream
New client, close previous stream
already started
Something went wrong: connect ECONNREFUSED 172.18.0.3:8070
Too many retries, YOLO took more than 3 min to start, likely an error
0

Also the Ubuntu terminal does not work either …

My configuration is this:

NVIDIA Jetson Nano (Developer Kit Version)
   * Jetpack 4.3 [L4T 32.3.1]
   * NV Power Mode: 5W - Type: 1
   * jetson_stats.service: active
 - Libraries:
   * CUDA: 10.0.326
   * cuDNN: 7.6.3.28
   * TensorRT: 6.0.1.10
   * Visionworks: 1.6.0.500n
   * OpenCV: 4.1.1 compiled CUDA: NO
   * VPI: 0.1.0
   * Vulkan: 1.1.70

I am working so that a daemon every x minutes uploads the data to a server and clears the logs, so that ODC is not loaded with much information.

You recommend that I change Jetson, or use another tool. And is ODC going to have a version soon?

Greetings and thank you

[tdurand]

On the other hand, the memory of the Jetson is increasing little by little and after about four hours it saturates, causing it to reboot. My question is if the tracker information is in memory (I imagine that to make the heatmap) and if there is a way to clean it or just do not save it (it is an idea).

No the tracker information isn’t kept in memory… we just keep the past 300 frames, you can config that in the config.json

That said… there may be still some memory leak but I don’t think it comes from OpenDataCam node.js code… can you monitor the details of the memory use and see which process cause memory increase ? Might be darknet we use to have problems with it.

In bash try to use some tool to monitor memory like htop

One idea that comes to my mind is to query the “status” to get the temperature/cpu/memory of the ODC, or is this a crazy idea?

I guess you should query directly the jetson : https://github.com/rbonghi/jetson_stats …

Above all I want to emphasize thanks for your help and comments, and that at no time I want to take advantage of your work, and that you can count on my collaboration.

Many thanks, appreciated, also credits goes to @vsaw that was lately more active than me on support 👏 .

The project I am developing is quite simple, as it is about counting bicycles for a few hours a day. I’ve been running it for four days in real time

We are super keen to here about the different usage of OpenDataCam , would be great if you can share a picture and detail a bit you use case here: https://github.com/opendatacam/opendatacam/issues/84 if you feel like it. For v4 release want to showcase the diversity of projects that can benefit of OpenDataCam.

[vsaw]

@willytrueba Acking what @tdurand just said, this will not be solved by a bigger Jetson. My guess is there are two things here that we need to look at

1. Power Mode: 5W Power Mode lowers CPU freq etc. so it could be that Darknet is just taking to long to start in 5W mode. Try going to 10W mode following these instructions: https://www.jetsonhacks.com/2019/04/10/jetson-nano-use-more-power/
2. The message essentially means that the Darknet process could not be started. Besides power this could be caused by a flacky video device. Can you confirm that the USB-Camera is plugged in and working outside of ODC?