Oct 1 Devlog 7 Final

Added the monitoring and Deployed on RDK X5

• added the data storage feature
• now all the images and the data stores in the data folder i used .json file insted of sql because i knew .json lil bit i can understand it thats why i made storage system with that
• stored image also showed on the monitor dashboard
• in the monitor dashboard it have live feed of the detection also
• the live dashboard is local hosted on rdk x5

Deployed on RDK X5 Board

• i deployed the program using ssh function wirelessly with out any wired connection for the camera reference we have camera feed on dashboard
• the dashboard is hosted on rdk x5 using local port
• other all the things are similar to pc so nothing much to say but in the rdk we have lesser fps

Future of this project

• i will create more accurate program for more precise detection
• i will design cad model for mount of the helmet
• i will use rdkx5 local 10 AI tops to process this

i will continue this project in next event and complete it

(I uploaded a Youtube demo video also for this project )

Oct 1 Devlog 6

Added detection in webcam feed

• as per now i have to select image to run detection on it then its showing results
• now i added same detection video feed its detects real time almost correct (took some help of ai in it apt i dont have time 😭)
• detection works smooth in video feed also its shows as fps , angle and predicted reading in scale of 1-10
• i faced like like problem in lock in the circle and needle so i took some ai help to fix it
• it added looked in config in which it will locked in for fixed amount as configured after that it will redo detection
• it solve problem of unstable detection because detection is locked

future task

• to store this readings and pictures and save it
• run this whole thing on rdk

Oct 31 Devlog 5

Fixed the out of range problem in detection

• as per now i program i which it detects the gauge and calculates the needle angle and gives the reading between 1-10 bars (not totally accurate)
• so in last devlog i talked about i am having issues in the some images its showing out of range
• i found a problem which is a calculation issue by wraparound by model0 360 degree calculation (i took help of ai in it)
• then fixed the minimum angle boudary to handle minor detection errors

30 Oct Devlog 4

added reading with angle detection

• previously i calculated the angle of the needle from the image
• now i calculated the approx. reading from that angle of needle in the circle
• like i calculated that by cutting the bottom area and and calculating from start point to end point its predict reading based on that
• Yes its bad calculation but i tried multiple method but i have ocr the image first to get the numbers on gauge that makes the process slow
• so for now its only works on gauge which i have config and one at a time according to config / manual calibration
• only its give value in between 1 to 10 bar which can be incorrect but you have to calibrate it according to gauge type/spec

future task for now

• as per now i cant make it detecting real reading due to ocr need for that , i will only have 1-10 scale for now
• and solve that out of range problem (image given)
• i will now focus on integrating in video feed and calculating in real time

26 Oct Devlog 3

added angle detection

• i added angle detection in the gauge now its detects angle of the needle by uploading photo of it
• i added one more feature to save the results from which i can store data
• i added the maximum and minimum detection by capturing darkest spot on gauge on both edges

tried to increase accuracy

• so i tried to create a new program and increase some accuracy or precision but it did worked it got worst and worst so decided to scarp that

• now i am focusing on angle detection which almost working but doing more work on it

Oct 21 Devlog 2

detecting the circle in the gauge

• i have fixed the fps issue previously and getting stable fps to perform my task through a webcam its mentioned in first devlog so i moved on to next part

• so after that i started working on detecting gauge and its main pin

• first i started with detecting the circle and middle point of the gauge so i used gauge image for that for now and then i apply same algorithm in video input

• so i understanded the geometry of circle to detect circle in the gauge

• then i get to know that we have prepare the image first for the detection in which we apply grayscale filter and blur to image , blur to reduce the noise

• after that preparing i used the “cv2.HoughCircles()” library of open cv this library search the circular shapes in images it have multiple parameter we can tweak to get desirable outcome

• after using that library to detect circle i draw the circle in the image and displayed the output in terminal

• after detecting the main circle in gauge finding the pin mid point is easy just put an dot on middle of circle(repented by red)

Long Part Detecting the needle

• to detect the main gauge needle we have to mask the dial gray scale the region and using edge detection to highlight the boundaries and line (photos added)
• then identify the candidate line around the gauge so they all not mix-up (only detects straight lines )
• then i visualized the lines
• i faced like problem in unpacking this detected line coordinates because it is structed as a single integer
• so i lookup for solution and come accross it we have to use “fatten()” it coverts arrays into one-dimensional array i didnt get it that much but its like assigning multiple coordinate at single point
• and i got lines on the images but all are not filtered too many line the out put is usable
• so we have to use filters like in the sence a set rules which line can pass through which are not
• i mused midpoint formula to calculate the closest line to the mid point to detect the needle
• after that i used angle calculation too
• after that i used all of this filtration and all that and combine it together and got the output
• in last now its detecting the needle and dial of the gauge given in input

(some time is inflated due to some error in hackatime i hope it solve asap it tracked my unrelated grokbot session time in this project )

Oct 18 Devlog 1

Starting up Project Testing all things setting up

• Making a inspection helmet for industry use where they use it for gauge reading and storing that data and its used for live monitoring

What did i do today

• i checked the new webcam i choose this cam because its best for my use , got its feed on window using python

• decided using open cv library of python to track the gauge data live

• learned you to add images in opencv window and add filters to them learned how to apply different types of filters

Problems -

• i faced from in getting stable fps in video feed using web cam so i tried multiple technique then i figured out by lowering the resolution and changing the video format give me output that i want

• so set the resolution to 640X480 and format to MJPG

in future

• now i am testing in my pc then i shift to my rdk x5 board and use it as computation board and use its npu to render output and attach this to a helmet so works can use this