Running run_classifier_continuous() over accellerometer data

ivan2an · February 23, 2026, 12:25pm

Hi,
I generated C++ library for CM4F generated to recognize 3 gestures - idle, updown, wave (followed the example from the youtube video). I integrated that successfully on my custom board.
When I think about gesture recognition it should process data not on a frame level but rather continuously. run_classifier_continuous() seems to provide what I was looking for.
I modified my application to initialize the classifier and then call run_classifier_continuous() but I got EI_IMPULSE_DSP_ERROR(-5).
Are there examples of how to use run_classifier_continuous()? Or is the continuous mode not suitable for gesture recognition?

AIWintermuteAI · March 2, 2026, 9:12am

Hi, @ivan2an !
Yes, run_classifier_continuous is hard coded to audio data - I dug out one of my own old examples for light sensor data, that is more similar to accelerometer.

github.com

Seeed-Studio/Seeed_Arduino_Sketchbook/blob/4f39ac4578267b5eeb1b6093f38b4efd02128c83/examples/WioTerminal_TinyML_1_Intro/Wio_Terminal_EI_Rock_Paper_Scissors_Continious/Wio_Terminal_EI_Rock_Paper_Scissors_Continious.ino#L167


      
          *
          * @param[in]  debug  Get debug info if true
          */
          static void read_data(void* pvParameters)
          {
              while (1) {
                  // Determine the next tick (and then sleep later)
                  uint64_t next_tick = micros() + (EI_CLASSIFIER_INTERVAL_MS * 1000);
          
          
        // roll the buffer -3 points so we can overwrite the last one
                  numpy::roll(buffer, EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE, -1);
          
          
        buffer[EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE - 1] = analogRead(WIO_LIGHT);
          
          
        // and wait for next tick
                  uint64_t time_to_wait = next_tick - micros();
                  delay((int)floor((float)time_to_wait / 1000.0f));
                  delayMicroseconds(time_to_wait % 1000);
              }
          }

It is for FreeRTOS based device,so you’ll need to adapt to what you have. But from what I remember, the crux of how to make it work was around numpy roll function.