Affordable and simple indoor positioning system using multiple LD2450 mmWave radar sensors

This is a work-in-progress project, born from my original concept and rapidly implemented. It’s my hope that you find inspiration in this endeavor and feel motivated to enhance and evolve it further. I look forward to seeing how the community can transform this into an even more extraordinary and marvelous project.
(*chatgpt :wink: )

The goal is to automate based on predicting user behaviors.

Assumptions:

  • Calculating motion vectors to determine the direction in which a person is moving.
  • Predicting behaviors based on tracking the trajectory of movement, likely using a neural network.
  • The ability to create zones and triggers, such as circles or polygons.
  • The option to expand the system with sensors monitoring breathing, heart rate, and heartbeat, which can be particularly useful in caring for infants and the elderly.
  • Minimizing delays in the system’s response.
  • A simple tracking algorithm for assigning unique identifiers (IDs) to tracked individuals.

Challenges and Problems:

  • Calibration of sensors relative to each other is time-consuming and cumbersome.
  • Blind spots within the sensor range can lead to calibration loss.
  • Detectors sometimes return distorted data, resulting in a “warped” Cartesian system.

Solutions:

  • Using a custom UART controller, bypassing HA. The controller serializes data into JSON format to reduce database load and minimize delays. Only then do the triggers return information to the sensors in HA.
  • Combining ld2410+ld2450, as seen in the attached photo, increases the efficiency of detecting the presence of people and static objects.
  • To limit noise, points are filtered only to the area they cover (defined by the user). The idea is to prevent reflections from, for example, the front of a refrigerator from misleading the system.

How does it work?

  • For each sensor, I adjust the Cartesian system by rotating and shifting it.
  • Data from the sensors are streamed.
  • For each returned point, I search for the nearest point with an assigned ID. This process takes into account the speed and direction of the object’s movement.
  • Checking if an object is within a polygon or within a point’s radius, then sending a trigger.

The LD2461s are arriving in January, and it’s going to be fun :smiley:



5 Likes

Did you face any heating issue?

I didn’t notice, but sometimes the LD2450 hangs.

The most frustrating aspect is the inconsistency of the data. It feels like the sensor occasionally shifts X and Y by a random value. The points literally float, but this happens at fairly long intervals, such as an hour or a day (at least that’s when I notice it).

1 Like

Fascinating. Looking forward to updates.

I have some Inovelli mmwave switches coming hopefully soon, I’m anxious to see how that does but will want to augment that with some extra stuff so this is an interesting thread to keep up on for that.

How did they turn out? Checking them out, but would like to hear from people who are already using them.

They haven’t shipped yet, they are on the second beta.

Hello, I’m very excited about your project, can you check your inbox?

@CO_4X4 @odwide

2461? are useless, my bet is that there was idea to connect two S5KM312CL as master/slave (you can connect up to 4), and it should work nicely however there is something wrong with firmware and nothing of it is working as expected, there is super ghosting, literally ghost are walking in empty room, targets are duplicated, reflected etc.

You can see here my few experiments, but lastly I received TI IWR6843AOP, and man, it is expensive but support, software, minimal lag - it is top tier.

About LD2450 you can look here too: HLK-LD2450 Initial experiments to connect to HomeAssistant - #470 by wojciech6789

@tieppm1 there is nothing in inbox from you

1 Like

Please check chat