Gilt_tracking_dataset / code /yolo_detection.py
anilbhujel's picture
Updated code
52d6f05
import cv2
import numpy as np
from pathlib import Path
from ultralytics import YOLO
from collections import defaultdict
import time
import json
# Configuration
INPUT_VIDEOS_DIR = "datasets/usplf/tracking/pen2_cam2" # Directory containing input videos
OUTPUT_DIR = "datasets/usplf/tracking/detected_json/pen2_cam2"
# POLYGON_VERTICES = np.array([[89,139], [325,57], [594, 6], [1128, 29], [1129, 364], [1031, 717],[509, 653],[287, 583], [100, 506],[74, 321]]) # Pen2 Cam1 polygon coordinates
POLYGON_VERTICES = np.array([[179, 3], [844, 8], [1151, 137], [1151, 316], [1135, 486], [995, 531],[801, 592],[278, 711], [167, 325]]) # Pen2 Cam2 polygon coordinates
CONF_THRESH = 0.55 # Confidence threshold
# TRACKER_CONFIG = "custom_bytetrack.yaml" # Built-in tracker config
# Visualization settings
SHOW_MASK_OVERLAY = True
MASK_ALPHA = 0.3 # Transparency for polygon mask
BOX_COLOR = (0, 255, 0) # Green
TEXT_COLOR = (255, 255, 255) # White
FONT_SCALE = 0.8
THICKNESS = 2
def create_mask(frame_shape):
mask = np.zeros(frame_shape[:2], dtype=np.uint8)
cv2.fillPoly(mask, [POLYGON_VERTICES], 255)
return mask
def draw_visuals_detection(frame, mask, detections, frame_count, fps):
if SHOW_MASK_OVERLAY:
overlay = frame.copy()
cv2.fillPoly(overlay, [POLYGON_VERTICES], (0, 100, 0))
cv2.addWeighted(overlay, MASK_ALPHA, frame, 1 - MASK_ALPHA, 0, frame)
for det in detections:
x, y, w, h = det['bbox']
conf = det['confidence']
# Draw bounding box
cv2.rectangle(
frame,
(int(x - w / 2), int(y - h / 2)),
(int(x + w / 2), int(y + h / 2)),
BOX_COLOR, THICKNESS
)
# Display confidence
cv2.putText(frame, f"{conf:.2f}",
(int(x - w / 2), int(y - h / 2) - 10),
cv2.FONT_HERSHEY_SIMPLEX, FONT_SCALE, TEXT_COLOR, THICKNESS)
# Overlay info
cv2.putText(frame, f"Frame: {frame_count}", (10, 30),
cv2.FONT_HERSHEY_SIMPLEX, FONT_SCALE, TEXT_COLOR, THICKNESS)
cv2.putText(frame, f"FPS: {fps:.1f}", (10, 60),
cv2.FONT_HERSHEY_SIMPLEX, FONT_SCALE, TEXT_COLOR, THICKNESS)
cv2.putText(frame, f"Pigs: {len(detections)}", (10, 90),
cv2.FONT_HERSHEY_SIMPLEX, FONT_SCALE, TEXT_COLOR, THICKNESS)
return frame
def process_video(video_path, output_path):
model = YOLO("trained_model_weight/pig_detect/yolo/pig_detect_pen2_best.pt")
cap = cv2.VideoCapture(str(video_path))
frame_count = 0
results_list = []
fps_history = []
# Video properties
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
mask = create_mask((frame_height, frame_width))
win_name = f"Pig Detection - {video_path.name}"
cv2.namedWindow(win_name, cv2.WINDOW_NORMAL)
while cap.isOpened():
start_time = time.time()
success, frame = cap.read()
if not success:
break
masked_frame = cv2.bitwise_and(frame, frame, mask=mask)
# Detection instead of tracking
results = model.predict(
masked_frame,
conf=CONF_THRESH,
verbose=False
)
detections = []
if results[0].boxes is not None:
boxes = results[0].boxes.xywh.cpu().numpy()
scores = results[0].boxes.conf.cpu().numpy()
for box, score in zip(boxes, scores):
x, y, w, h = box
detections.append({"confidence": float(score), "bbox": [x, y, w, h]})
results_list.append({
"frame_id": frame_count,
"frame_width": frame_width,
"frame_height": frame_height,
"confidence": float(score),
"bbox": [float(x), float(y), float(w), float(h)],
"area": float(w * h)
})
# FPS calculation
processing_time = time.time() - start_time
fps = 1 / processing_time
fps_history.append(fps)
if len(fps_history) > 10:
fps = np.mean(fps_history[-10:])
# Draw visuals
display_frame = draw_visuals_detection(frame.copy(), mask, detections, frame_count, fps)
cv2.imshow(win_name, display_frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
frame_count += 1
cap.release()
cv2.destroyWindow(win_name)
# Save JSON
output_file = output_path / f"{video_path.stem}_detection.json"
with open(output_file, 'w') as f:
json.dump(results_list, f, indent=2)
if __name__ == "__main__":
input_dir = Path(INPUT_VIDEOS_DIR)
output_dir = Path(OUTPUT_DIR)
output_dir.mkdir(exist_ok=True)
for video_file in input_dir.glob("*.mp4"):
print(f"Processing {video_file.name}...")
process_video(video_file, output_dir)
cv2.destroyAllWindows()