CODEWITHSUNDEEP
pythonlinuxraspberry-pi5debian-bookwormpicamera2
Bsaka
Bsaka

Reputation: 1

Error while trying to use the picam2 library: module 'kms' has no attribute 'PixelFormat'

I have been facing an issue. I have a project, where I have to use facial recognition from insightface to detect know people. Hardware-wise, I have been given a Raspberry pi 5 and two picams (v1.3, v2.1). I found out that there is an issue with the picam 2 not being detected by the system on pi5, so I am using the version 1.3. However, the issue I have been facing is the constant error AttributeError: module 'kms' has no attribute 'PixelFormat'.

To troubleshoot it, I reinstalled the picam2 library, verified that I do not have clashing versions, tried using no preview. No matter what I try, the error occurs.

import cv2
from picamera2 import Picamera2
import numpy as np
import pickle
from insightface.app import *
import Custom_Buzzer
from gpiozero import Device
from gpiozero.pins.lgpio import LGPIOFactory
Device.pin_factory = LGPIOFactory()
picam2 = Picamera2(preview=None)
picam2.configure(picam2.create_still_configuration(main={"format": "RGB888", "size": (640, 480)}))
picam2.start()

# Load the known face embeddings
with open("known_faces.pkl", "rb") as f:
    known_face_encodings, known_face_names = pickle.load(f)

# Initialize the FaceAnalysis app
app = FaceAnalysis()
app.prepare(ctx_id=-1, det_size=(640, 640))

buzzer = Custom_Buzzer.create_buzzer(16)

while True:
    # Capture frame-by-frame
    frame = picam2.capture_array()
    
    rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

    # Perform face detection
    faces = app.get(rgb_frame)

    for face in faces:
        # Compute the face embedding
        face_embedding = face.normed_embedding

        # Compare against known faces
        distances = np.linalg.norm(np.array(known_face_encodings) - face_embedding, axis=1)
        min_distance_index = np.argmin(distances)
        min_distance = distances[min_distance_index]

        # Set a threshold for the distance to consider it a match
        threshold = 1.0
        if min_distance < threshold:
            name = known_face_names[min_distance_index]
            Custom_Buzzer.initialize_buzzer(buzzer, 2)
        else:
            name = "Unknown"
            Custom_Buzzer.initialize_buzzer(buzzer, 5)

Upvotes: 0

Views: 103

Answers (1)

Bsaka
Bsaka

Reputation: 1

This is not really a solution, as I believe an issue exists with the libraries currently, but I found a workaround, by basically saving short videos with libcamera which turns out to be functional, and then parsing that as input for analysis for the opencv

Upvotes: 0

Related Questions