This is an old revision of the document!

Thermal Imaging

Thermal camera transmits udp packets to a windows application.

TODO

~~capture udp packets~~
~~analyze for handshake and frame~~
~~cross platform image stream decoding~~
find camera's conversion formula or lookup table OR calibrate
- ~~grab calibration data~~
- get min max value and saturation points
- normalize, colorize and manipulate 16bit image
Live streaming
web interface
Mobilility (battary, SOB)

flow

we recorded a raw ethernet packet stream between the camera (source) and the windows (target) pc

bootloginplay.zip - pcap file recording of camera boot, stream request follwed by ~100 frames of thermal video. we used wireshark to filter and capture the handshake.

we noticed the stream gets initiated by a short udp packet from the target
we saved that packet and replayed it to the source using scapy lib for python.

#replay the "trigger" packet. 
#this packets will start the source broadcasting its packets. 
 
import base64
from scapy.all import *
 
# Base64 encoded packet data 
encoded_packet = "////////AAFsWfAKCABFAAA4KB0AAIARkEfAqAABwKgA/x+bH5wA2QAAASABgBtAACAAAAAAAAAADwAAAAEAAAEAACArAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAP//////////////////////////////////////////AAAAAAAAAAIBAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA"
 
 
# Decode the Base64 encoded packet
decoded_packet = base64.b64decode(encoded_packet)
 
# Load packet with Scapy
packet = Ether(decoded_packet)
 
# (packet)
sendp(packet, iface="Ethernet")

this then started a network stream
the data stream was ordered and looked clean.
every frame starts with the following.

T=(-1.665884e-08)*X^4+(1.347094e-05)*X^3+(-4.396264e-03)*X^2+(9.506939e-01)*X+(-6.353247e+01)

that got our interest in that packet but from then we noticed each has a sequential header that we can follow.

we then decoded to video > https://git.telavivmakers.space/tami/thermalcam_decoder

frame composed of 32*6930b packets
384×288 px
16bit grayscale
first 520b overwritten

here is a video (mp4,h264,8bpp) and an image (png,raw,16bpp)

raw data of single frame, in pcap format (TODO: not the same data as the png above.), frame2.7z

calibrate

~~we don't know the camera's conversion formula or lookup table~~
the spec mentions 8-14µm

According to Wien's Displacement Law, objects at different temperatures emit peak radiation at different wavelengths. The range of 8-14 µm is well-suited for detecting temperatures commonly encountered in industrial and environmental applications, typically ranging from about -20°C to 500°C or more. This range is sensitive enough to detect the radiation emitted by most objects under normal conditions.

i asked gpt if the formula we found in the first udp packet makes sense as a lookup table. he said no

then decided to capture a calibration sequence using our ANET printer hotbed and tip as a commendable temperature source.
set the hotbed to 100c (but its hardcoded to 90C tops on the merlin firmware) and 200deg for the hotend tip.
then “showed” it a piece of ice from the fridge. all measurements verified using a infrared thermometer. 90C,200C and 0C.

https://archive.org/details/heatbed90_hotend_200_ice0.7z
- first attempt, 5Gb (compressed to 800Mb)
- long 15min take include ramp up (shown below at x30 speed)

and revisiting that formula, what do you know!? it is legit…

check this rust thingy out (more at the repo)

    let frame = data
        .iter()
        .copied()
        .map(|x| {
            let x: f64 = x.into();
            let x = x / 256.0;
            ((-1.665884e-08) * x.powf(4.)
                + (1.347094e-05) * x.powf(3.)
                + (-4.396264e-03) * x.powf(2.)
                + (9.506939e-01) * x
                + (-6.353247e+01)) as u8
        })
        .collect::<Vec<u8>>();