This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

How to implement automatic clustering ?

Other Parts Discussed in Thread: IWR1642

Hello,
My projet is to build an autonomous device, based on TI IWR mmwave sensor. It would by a 15 meters range radar, capable of detecting cars and determining if they are stopped or not. I would like it simple to configure and to use.
Here is the use-case :
1. The user choose a number N of different situations (scenes) he wants to recognize. Typically between 2 and 5. This is the only parameter of the device.
2. The user switches the device in "learning mode". In this mode, the device build a classification method using unsupervised learning algorithm (k-means f. example).
3. The device has 5 LEDs to ouput the real-time classification done by the device. The user monitors how the learning process is going on, and check if it fits his needs (if not : change the location of the radar, or change N parameter).
4. When the user considers that the learning process is successful, he switches the device in "clustering mode". The device does not learn anymore. It observes the scene, calculates its category, and lights the appropriate LED.
5. The device counts the number of occurences for each category, during a fixed time lap (5 minutes f. example). It sends the results to another device using I2C, for feeding a database.

My questions are :
- Which mmwave sensor is the most suitable to do this ?
- How to implement this with a standalone sensor, without adding any other chip to run the learning process ?
- Any suggestion or idea to make it ?

Thank you !

Halden

  • Hello Halden,
    At a first look the IWR1642 sensor might be best suited for the application you describe above. I have assigned this thread to an expert who would get back to you on more details on clustering etc.

    Regards,
    Vivek
  • Hello Halden,

    Based on thing you have listed implementation will span standard FMCW processing to get data point cloud information and then further cluster/track these points to make high level decisions.

    Please look at the Out of Box demo for IWR1642 device which shows how points of object and data is collected from scene.

    Example usage of identical Traffic Monitoring scenario using IWR162 EVM is discussed here : Robust traffic and intersection monitoring using millimeter wave sensors

    Above white paper discusses various sensing parameters for this application and tradeoffs of configuration on IWR1642 from standard processing perspective.

    There are various high level algorithms available which can be chosen to tracking and clustering. It is up to the user to select the one appropriate for their application. IWR1642 has DSP Core which can utilized of such algorithms.

    Hope this gives you a direction to explore further and try out IWR1642 EVM.

    Thank you,

    Vaibhav

  • Closing due to inactivity. If you have additional questions, please start a new thread.