VisionScout / video_processor.py
DawnC's picture
Add new feature "Video Process" and fix format issue
c0fe80d verified
raw
history blame
18.7 kB
import cv2
import os
import tempfile
import uuid
from PIL import Image
import numpy as np
from typing import Dict, List, Tuple, Any, Optional
import time
from collections import defaultdict
from image_processor import ImageProcessor
from evaluation_metrics import EvaluationMetrics
from scene_analyzer import SceneAnalyzer
from detection_model import DetectionModel
class VideoProcessor:
"""
Handles the processing of video files, including object detection
and scene analysis on selected frames.
"""
def __init__(self, image_processor: ImageProcessor):
"""
Initializes the VideoProcessor.
Args:
image_processor (ImageProcessor): An initialized ImageProcessor instance.
"""
self.image_processor = image_processor
def process_video_file(self,
video_path: str,
model_name: str,
confidence_threshold: float,
process_interval: int = 5,
scene_desc_interval_sec: int = 3) -> Tuple[Optional[str], str, Dict]:
"""
Processes an uploaded video file, performs detection and periodic scene analysis,
and returns the path to the annotated output video file along with a summary.
Args:
video_path (str): Path to the input video file.
model_name (str): Name of the YOLO model to use.
confidence_threshold (float): Confidence threshold for object detection.
process_interval (int): Process every Nth frame. Defaults to 5.
scene_desc_interval_sec (int): Update scene description every N seconds. Defaults to 3.
Returns:
Tuple[Optional[str], str, Dict]: (Path to output video or None, Summary text, Statistics dictionary)
"""
if not video_path or not os.path.exists(video_path):
print(f"Error: Video file not found at {video_path}")
return None, "Error: Video file not found.", {}
print(f"Starting video processing for: {video_path}")
start_time = time.time()
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
print(f"Error: Could not open video file {video_path}")
return None, "Error opening video file.", {}
# Get video properties
fps = cap.get(cv2.CAP_PROP_FPS)
if fps <= 0: # Handle case where fps is not available or invalid
fps = 30 # Assume a default fps
print(f"Warning: Could not get valid FPS for video. Assuming {fps} FPS.")
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
total_frames_video = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
print(f"Video properties: {width}x{height} @ {fps:.2f} FPS, Total Frames: {total_frames_video}")
# Calculate description update interval in frames
description_update_interval_frames = int(fps * scene_desc_interval_sec)
if description_update_interval_frames < 1:
description_update_interval_frames = int(fps) # Update at least once per second if interval is too short
object_trackers = {} # 儲存ID與物體的映射
last_detected_objects = {} # 儲存上一次檢測到的物體資訊
next_object_id = 0 # 下一個可用的物體ID
tracking_threshold = 0.6 # 相同物體的IoU
object_colors = {} # 每個被追蹤的物體分配固定顏色
# Setup Output Video
output_filename = f"processed_{uuid.uuid4().hex}_{os.path.basename(video_path)}"
temp_dir = tempfile.gettempdir() # Use system's temp directory
output_path = os.path.join(temp_dir, output_filename)
# Ensure the output path has a compatible extension (like .mp4)
if not output_path.lower().endswith(('.mp4', '.avi', '.mov')):
output_path += ".mp4"
# Use 'mp4v' for MP4, common and well-supported
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))
if not out.isOpened():
print(f"Error: Could not open VideoWriter for path: {output_path}")
cap.release()
return None, f"Error creating output video file at {output_path}.", {}
print(f"Output video will be saved to: {output_path}")
frame_count = 0
processed_frame_count = 0
all_stats = [] # Store stats for each processed frame
summary_lines = []
last_description = "Analyzing scene..." # Initial description
frame_since_last_desc = description_update_interval_frames # Trigger analysis on first processed frame
try:
while True:
ret, frame = cap.read()
if not ret:
break # End of video
frame_count += 1
frame_since_last_desc += 1
current_frame_annotated = False # Flag if this frame was processed and annotated
# Process frame based on interval
if frame_count % process_interval == 0:
processed_frame_count += 1
print(f"Processing frame {frame_count}...")
current_frame_annotated = True
# Use ImageProcessor for single-frame tasks
# 1. Convert frame format BGR -> RGB -> PIL
try:
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_image = Image.fromarray(frame_rgb)
except Exception as e:
print(f"Error converting frame {frame_count}: {e}")
continue # Skip this frame
# 2. Get appropriate model instance
# Confidence is passed from UI, model_name too
model_instance = self.image_processor.get_model_instance(model_name, confidence_threshold)
if not model_instance or not model_instance.is_model_loaded:
print(f"Error: Model {model_name} not loaded. Skipping frame {frame_count}.")
# Draw basic frame without annotation
cv2.putText(frame, f"Scene: {last_description[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 3, cv2.LINE_AA)
cv2.putText(frame, f"Scene: {last_description[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
out.write(frame)
continue
# 3. Perform detection
detection_result = model_instance.detect(pil_image) # Use PIL image
current_description_for_frame = last_description # Default to last known description
scene_analysis_result = None
stats = {}
if detection_result and hasattr(detection_result, 'boxes') and len(detection_result.boxes) > 0:
# Ensure SceneAnalyzer is ready within ImageProcessor
if not hasattr(self.image_processor, 'scene_analyzer') or self.image_processor.scene_analyzer is None:
print("Initializing SceneAnalyzer...")
# Pass class names from the current detection result
self.image_processor.scene_analyzer = SceneAnalyzer(class_names=detection_result.names)
elif self.image_processor.scene_analyzer.class_names is None:
# Update class names if they were missing
self.image_processor.scene_analyzer.class_names = detection_result.names
if hasattr(self.image_processor.scene_analyzer, 'spatial_analyzer'):
self.image_processor.scene_analyzer.spatial_analyzer.class_names = detection_result.names
# 4. Perform Scene Analysis (periodically)
if frame_since_last_desc >= description_update_interval_frames:
print(f"Analyzing scene at frame {frame_count} (threshold: {description_update_interval_frames} frames)...")
# Pass lighting_info=None for now, as it's disabled for performance
scene_analysis_result = self.image_processor.analyze_scene(detection_result, lighting_info=None)
current_description_for_frame = scene_analysis_result.get("description", last_description)
last_description = current_description_for_frame # Cache the new description
frame_since_last_desc = 0 # Reset counter
# 5. Calculate Statistics for this frame
stats = EvaluationMetrics.calculate_basic_stats(detection_result)
stats['frame_number'] = frame_count # Add frame number to stats
all_stats.append(stats)
# 6. Draw annotations
names = detection_result.names
boxes = detection_result.boxes.xyxy.cpu().numpy()
classes = detection_result.boxes.cls.cpu().numpy().astype(int)
confs = detection_result.boxes.conf.cpu().numpy()
def calculate_iou(box1, box2):
"""Calculate Intersection IOU value"""
x1_1, y1_1, x2_1, y2_1 = box1
x1_2, y1_2, x2_2, y2_2 = box2
xi1 = max(x1_1, x1_2)
yi1 = max(y1_1, y1_2)
xi2 = min(x2_1, x2_2)
yi2 = min(y2_1, y2_2)
inter_area = max(0, xi2 - xi1) * max(0, yi2 - yi1)
box1_area = (x2_1 - x1_1) * (y2_1 - y1_1)
box2_area = (x2_2 - x1_2) * (y2_2 - y1_2)
union_area = box1_area + box2_area - inter_area
return inter_area / union_area if union_area > 0 else 0
# 處理當前幀中的所有檢測
current_detected_objects = {}
for box, cls_id, conf in zip(boxes, classes, confs):
x1, y1, x2, y2 = map(int, box)
# 查找最匹配的已追蹤物體
best_match_id = None
best_match_iou = 0
for obj_id, (old_box, old_cls_id, _) in last_detected_objects.items():
if old_cls_id == cls_id: # 同一類別才比較
iou = calculate_iou(box, old_box)
if iou > tracking_threshold and iou > best_match_iou:
best_match_id = obj_id
best_match_iou = iou
# 如果找到匹配,使用現有ID;否則分配新ID
if best_match_id is not None:
obj_id = best_match_id
else:
obj_id = next_object_id
next_object_id += 1
# 為新物體分配固定顏色 - 使用更明顯的顏色
# 使用更明顯的顏色,避免白色
bright_colors = [
(0, 0, 255), # red
(0, 255, 0), # green
(255, 0, 0), # blue
(0, 255, 255), # yellow
(255, 0, 255), # purple
(255, 128, 0), # orange
(128, 0, 255) # purple
]
object_colors[obj_id] = bright_colors[obj_id % len(bright_colors)]
# update tracking info
current_detected_objects[obj_id] = (box, cls_id, conf)
color = object_colors.get(obj_id, (0, 255, 0)) # default is green
label = f"{names.get(cls_id, 'Unknown')}-{obj_id}: {conf:.2f}"
# 平滑化邊界框:如果是已知物體,與上一幀位置平均
if obj_id in last_detected_objects:
old_box, _, _ = last_detected_objects[obj_id]
old_x1, old_y1, old_x2, old_y2 = map(int, old_box)
# 平滑係數
alpha = 0.7 # current weight
beta = 0.3 # history weight
x1 = int(alpha * x1 + beta * old_x1)
y1 = int(alpha * y1 + beta * old_y1)
x2 = int(alpha * x2 + beta * old_x2)
y2 = int(alpha * y2 + beta * old_y2)
# draw box and label
cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
# add text
(w, h), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 2)
cv2.rectangle(frame, (x1, y1 - h - 10), (x1 + w, y1 - 10), color, -1)
cv2.putText(frame, label, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1, cv2.LINE_AA)
# update tracking info
last_detected_objects = current_detected_objects.copy()
# Draw the current scene description on the frame
cv2.putText(frame, f"Scene: {current_description_for_frame[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 3, cv2.LINE_AA) # Black outline
cv2.putText(frame, f"Scene: {current_description_for_frame[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA) # White text
# Write the frame (annotated or original) to the output video
# Draw last known description if this frame wasn't processed
if not current_frame_annotated:
cv2.putText(frame, f"Scene: {last_description[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 0), 3, cv2.LINE_AA)
cv2.putText(frame, f"Scene: {last_description[:80]}...", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2, cv2.LINE_AA)
out.write(frame) # Write frame to output file
except Exception as e:
print(f"Error during video processing loop for {video_path}: {e}")
import traceback
traceback.print_exc()
summary_lines.append(f"An error occurred during processing: {e}")
finally:
# Release resources
cap.release()
out.release()
print(f"Video processing finished. Resources released. Output path: {output_path}")
if not os.path.exists(output_path) or os.path.getsize(output_path) == 0:
print(f"Error: Output video file was not created or is empty at {output_path}")
summary_lines.append("Error: Failed to create output video.")
output_path = None
end_time = time.time()
processing_time = end_time - start_time
summary_lines.insert(0, f"Finished processing in {processing_time:.2f} seconds.")
summary_lines.insert(1, f"Processed {processed_frame_count} frames out of {frame_count} (interval: {process_interval} frames).")
summary_lines.insert(2, f"Scene description updated approximately every {scene_desc_interval_sec} seconds.")
# Generate Aggregate Statistics
aggregated_stats = {
"total_frames_read": frame_count,
"total_frames_processed": processed_frame_count,
"avg_objects_per_processed_frame": 0, # Calculate below
"cumulative_detections": {}, # Total times each class was detected
"max_concurrent_detections": {} # Max count of each class in a single processed frame
}
object_cumulative_counts = {}
object_max_concurrent_counts = {} # Store the max count found for each object type
total_detected_in_processed = 0
# Iterate through stats collected from each processed frame
for frame_stats in all_stats:
total_objects_in_frame = frame_stats.get("total_objects", 0)
total_detected_in_processed += total_objects_in_frame
# Iterate through object classes detected in this frame
for obj_name, obj_data in frame_stats.get("class_statistics", {}).items():
count_in_frame = obj_data.get("count", 0)
# Cumulative count
if obj_name not in object_cumulative_counts:
object_cumulative_counts[obj_name] = 0
object_cumulative_counts[obj_name] += count_in_frame
# Max concurrent count
if obj_name not in object_max_concurrent_counts:
object_max_concurrent_counts[obj_name] = 0
# Update the max count if the current frame's count is higher
object_max_concurrent_counts[obj_name] = max(object_max_concurrent_counts[obj_name], count_in_frame)
# Add sorted results to the final dictionary
aggregated_stats["cumulative_detections"] = dict(sorted(object_cumulative_counts.items(), key=lambda item: item[1], reverse=True))
aggregated_stats["max_concurrent_detections"] = dict(sorted(object_max_concurrent_counts.items(), key=lambda item: item[1], reverse=True))
# Calculate average objects per processed frame
if processed_frame_count > 0:
aggregated_stats["avg_objects_per_processed_frame"] = round(total_detected_in_processed / processed_frame_count, 2)
summary_text = "\n".join(summary_lines)
print("Generated Summary:\n", summary_text)
print("Aggregated Stats (Revised):\n", aggregated_stats) # Print the revised stats
# Return the potentially updated output_path
return output_path, summary_text, aggregated_stats