OV7675 Camera from TinyML kit not capturing image properly

[robotkat11]

Hi,
I've started a course from HarvardX on EDX that introduces me to TinyML. One of the lessons includes using the Arduino TinyML kit which consists of the Arduino Nano 33 BLE Sense board and the OV7675 camera among other components. I've followed two tutorials on how to get started and managed to take a photo, the only issue is that the hexadecimal printout when converted looks saturated. I don't know if this is due to the code or something with the hardware itself and I can't find much else about the camera other than the EDX course and handout.

Here is an example image of what I am seeing.

The code compiled onto the board can be found in the Arduino IDE : File → Examples → Harvard_TinyMLx → test_camera

Here is the code as well

#include <Arduino_OV767X.h>
#include <mbed.h>

static mbed::DigitalIn button(p28);
#define PRESSED 0

int bytes_per_frame;
int bytes_per_pixel;

uint8_t data[160 * 120 * 2]; // QQVGA: 160x120 X 2 bytes per pixel (YUV422)

template <typename T>
inline T clamp_0_255(T x) {
  return std::max(std::min(x, static_cast<T>(255)), static_cast<T>(0));
}

inline void ycbcr422_rgb888(int32_t Y, int32_t Cb, int32_t Cr, uint8_t* out) {
  // Cb and Cr offset by 128
  Cr -= 128;
  Cb -= 128;

  // Convert YUV to RGB using standard conversion formulas
  out[0] = clamp_0_255(Y + (1.402 * Cr));                        // Red
  out[1] = clamp_0_255(Y - (0.344 * Cb) - (0.714 * Cr));         // Green
  out[2] = clamp_0_255(Y + (1.772 * Cb));                       // Blue
}

void setup() {
  Serial.begin(115200); // Correct baud rate
  while (!Serial);

  if (!Camera.begin(QQVGA, YUV422, 1)) {
    Serial.println("Failed to initialize camera!");
    while (1);
  }

  bytes_per_pixel = Camera.bytesPerPixel();
  bytes_per_frame = Camera.width() * Camera.height() * bytes_per_pixel;

  // Optionally enable a test pattern for debugging camera output
  // Camera.testPattern();
}

void loop() {
  if (button == PRESSED) {
    Camera.readFrame(data);
    uint8_t rgb888[3]; // Buffer to hold RGB values
    Serial.println("<image>");
    Serial.println(Camera.width());
    Serial.println(Camera.height());

    // Process the frame
    for (int i = 0; i < bytes_per_frame; i += bytes_per_pixel * 2) {
      // Read YUV422 data
      const int32_t Y0 = data[i];
      const int32_t Cr = data[i + 1];
      const int32_t Y1 = data[i + 2];
      const int32_t Cb = data[i + 3];

      // Convert first pixel (Y0, Cb, Cr)
      ycbcr422_rgb888(Y0, Cb, Cr, rgb888);
      Serial.print(rgb888[0]);
      Serial.print(" ");
      Serial.print(rgb888[1]);
      Serial.print(" ");
      Serial.println(rgb888[2]);

      // Convert second pixel (Y1, Cb, Cr)
      ycbcr422_rgb888(Y1, Cb, Cr, rgb888);
      Serial.print(rgb888[0]);
      Serial.print(" ");
      Serial.print(rgb888[1]);
      Serial.print(" ");
      Serial.println(rgb888[2]);
    }
    Serial.println("</image>");
  }
}

The code offered to convert hexadecimal to an image: colabs/4-2-12-OV7675ImageViewer.ipynb at master · tinyMLx/colabs · GitHub

The course and this handout show the similar instructions I followed: https://tinyml.seas.harvard.edu/assets/other/4D/22.03.11_Marcelo_Rovai_Handout.pdf

The course on EDX is titled: Deploying TinyML by HarvardX.

The conversion seems to work, so I believe it has to do with the compiled code for the Arduino, but I'm unsure what exactly would be the problem.

[sonofcy]

First enable the test pattern so you know if the camera is ok.

[robotkat11]

Apologies for the delay. I referenced the wrong code, but I incorporated Camera.testPattern(); in the Harvard ML tutorial. This was the image I received from running it once I converted it in the supplied notebook.

I found an article that mentions this test should output color bars. Here is where I found an example image (site: Tutorial Part-1 : TinyML with OV7670 Camera module! - Hackster.io) and this was their resulting image, though they used a different camera version from the one I have.

Seems like some of the bars for mine are cut off, but the color itself looks good, so would this possibly be an software issue then?

[sonofcy]

Without knowing what is expected, I have to assume that is normal.