import os import tempfile import numpy as np import streamlit as st from PIL import Image import tensorflow as tf import cv2 from pathlib import Path # ───────────────────────────────────────────── # PAGE CONFIG # ───────────────────────────────────────────── st.set_page_config( page_title="German Traffic Sign Recognition", page_icon="🚦", layout="wide", initial_sidebar_state="expanded", ) # ───────────────────────────────────────────── # CUSTOM CSS — dark industrial / road aesthetic # ───────────────────────────────────────────── st.markdown(""" """, unsafe_allow_html=True) # ───────────────────────────────────────────── # CLASS NAMES (43 classes — correct & verified) # ───────────────────────────────────────────── CLASS_NAMES = { 0: 'Speed limit (20km/h)', 1: 'Speed limit (30km/h)', 2: 'Speed limit (50km/h)', 3: 'Speed limit (60km/h)', 4: 'Speed limit (70km/h)', 5: 'Speed limit (80km/h)', 6: 'End of speed limit (80km/h)', 7: 'Speed limit (100km/h)', 8: 'Speed limit (120km/h)', 9: 'No passing', 10: 'No passing for vehicles over 3.5t', 11: 'Right-of-way at next intersection', 12: 'Priority road', 13: 'Yield', 14: 'Stop', 15: 'No vehicles', 16: 'Vehicles over 3.5t prohibited', 17: 'No entry', 18: 'General caution', 19: 'Dangerous curve to the left', 20: 'Dangerous curve to the right', 21: 'Double curve', 22: 'Bumpy road', 23: 'Slippery road', 24: 'Road narrows on the right', 25: 'Road work', 26: 'Traffic signals', 27: 'Pedestrians', 28: 'Children crossing', 29: 'Bicycles crossing', 30: 'Beware of ice/snow', 31: 'Wild animals crossing', 32: 'End of all speed and passing limits', 33: 'Turn right ahead', 34: 'Turn left ahead', 35: 'Ahead only', 36: 'Go straight or right', 37: 'Go straight or left', 38: 'Keep right', 39: 'Keep left', 40: 'Roundabout mandatory', 41: 'End of no passing', 42: 'End of no passing for vehicles over 3.5t', } # Category groupings for sidebar info CATEGORIES = { "🔴 Speed Limits": list(range(0, 9)), "⛔ Prohibition": [9,10,15,16,17], "⚠️ Warning": list(range(11,12)) + list(range(18,32)), "🔵 Mandatory": list(range(33,41)), "⬜ End of restriction": [6,32,41,42], "🛑 Priority": [12,13,14], } # ───────────────────────────────────────────── # MODEL LOADING # ───────────────────────────────────────────── BASE_DIR = Path(__file__).resolve().parent MODEL_PATH = BASE_DIR / "GermanTraffic-sıfırdanCNN-OS.keras" IMG_SIZE = (64, 64) # matches your training config @st.cache_resource(show_spinner=False) def load_model(): try: return tf.keras.models.load_model(MODEL_PATH) except Exception as e: st.error(f"Model yüklenemedi: {e}\n\nModel dosyasının app.py ile aynı klasörde olduğundan emin ol: `{MODEL_PATH}`") st.stop() # ───────────────────────────────────────────── # PREPROCESSING # ───────────────────────────────────────────── def preprocess(pil_img: Image.Image) -> np.ndarray: """Matches training pipeline exactly: RGB resize 64x64, /255.0""" img = pil_img.convert("RGB") img = img.resize(IMG_SIZE, Image.LANCZOS) x = np.array(img, dtype=np.float32) / 255.0 return np.expand_dims(x, axis=0) # (1, 64, 64, 3) def predict(pil_img: Image.Image): model = load_model() x = preprocess(pil_img) probs = model.predict(x, verbose=0)[0] # (43,) top5 = np.argsort(probs)[::-1][:5] return probs, top5 # ───────────────────────────────────────────── # VIDEO PROCESSING (from video_processor.py logic) # ───────────────────────────────────────────── def build_candidate_mask(frame: np.ndarray) -> np.ndarray: """ Combines THREE detection techniques from computer_vision.ipynb: 1. HSV color segmentation (Cell 18, 22) — isolates red & blue sign colors 2. Canny edge detection (Cell 27) — catches shape boundaries of any sign 3. Noise cleanup via dilate (Cell 41) — fills small gaps, merges nearby blobs Red → speed limits, prohibitions, warnings (majority of GTSRB classes) Blue → mandatory signs (keep right/left, roundabout, ahead only — classes 33-40) Canny→ catches yellow/white signs that HSV color mask misses """ hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # ── 1. Red mask (two HSV ranges because red wraps around 0°/180°) ────────── # Technique from computer_vision.ipynb Cell 18 & 22 red_mask = ( cv2.inRange(hsv, np.array([0, 70, 50]), np.array([10, 255, 255])) | cv2.inRange(hsv, np.array([170, 70, 50]), np.array([180,255, 255])) ) # ── 2. Blue mask (mandatory signs) ────────────────────────────────────────── # computer_vision.ipynb Cell 22 — channel isolation by hue blue_mask = cv2.inRange(hsv, np.array([100, 80, 50]), np.array([130, 255, 255])) # ── 3. Canny edge detection on grayscale ──────────────────────────────────── # computer_vision.ipynb Cell 27: Sobel/Laplacian/Canny comparison # GaussianBlur first (Cell 28 sketch function pattern) to reduce noise gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) blurred = cv2.GaussianBlur(gray, (5, 5), 0) canny = cv2.Canny(blurred, 50, 120) # same thresholds as Cell 27 # ── 4. Combine all three masks ─────────────────────────────────────────────── combined = cv2.bitwise_or(red_mask, blue_mask) combined = cv2.bitwise_or(combined, canny) # ── 5. Dilate to fill gaps (Cell 41 pattern: dilate + thresh) ─────────────── kernel = np.ones((3, 3), np.uint8) combined = cv2.dilate(combined, kernel, iterations=2) return combined def process_video_frames(video_path: str, conf_thresh: float = 0.60) -> str: """ Frame-by-frame traffic sign detection. Detection pipeline (techniques from computer_vision.ipynb): HSV red mask + HSV blue mask + Canny edges → dilate → findContours → CNN classify Returns path to processed output video. """ model = load_model() cap = cv2.VideoCapture(video_path) if not cap.isOpened(): return None fps = cap.get(cv2.CAP_PROP_FPS) or 20.0 width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) out_path = video_path.replace(".", "_processed.") fourcc = cv2.VideoWriter_fourcc(*"mp4v") out = cv2.VideoWriter(out_path, fourcc, fps, (width, height)) # ── Motion detection state (computer_vision.ipynb Cell 38-41) ─────────────── # absdiff between consecutive frames — only process regions with motion ret_prev, prev_frame = cap.read() if not ret_prev: cap.release() out.release() return None prev_gray = cv2.cvtColor(prev_frame, cv2.COLOR_BGR2GRAY) prev_blur = cv2.GaussianBlur(prev_gray, (21, 21), 0) while True: ret, frame = cap.read() if not ret: break # ── Motion filter (Cell 39 pattern) ───────────────────────────────────── curr_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) curr_blur = cv2.GaussianBlur(curr_gray, (21, 21), 0) diff = cv2.absdiff(prev_blur, curr_blur) motion_thresh = cv2.threshold(diff, 20, 255, cv2.THRESH_BINARY)[1] motion_pixels = np.sum(motion_thresh) / 255 prev_blur = curr_blur # advance frame # Skip static frames — if very few pixels changed, no new sign appeared # Threshold: at least 0.3% of frame must show motion if motion_pixels < (width * height * 0.003): out.write(frame) continue # ── Combined mask (HSV + Canny) ────────────────────────────────────────── candidate_mask = build_candidate_mask(frame) # ── Contour detection (computer_vision.ipynb Cell 42) ─────────────────── contours, _ = cv2.findContours( candidate_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE ) for cnt in contours: x, y, w, h = cv2.boundingRect(cnt) # Size filter: ignore tiny noise and full-frame false positives if w < 25 or h < 25 or w > width * 0.7 or h > height * 0.7: continue # Aspect ratio filter: traffic signs are roughly square (0.5 – 2.0) aspect = w / h if aspect < 0.4 or aspect > 2.5: continue roi = frame[y:y+h, x:x+w] roi_rgb = cv2.cvtColor(roi, cv2.COLOR_BGR2RGB) roi_pil = Image.fromarray(roi_rgb) roi_inp = preprocess(roi_pil) preds = model.predict(roi_inp, verbose=0)[0] class_id = int(np.argmax(preds)) confidence = float(np.max(preds)) if confidence >= conf_thresh: label = f"{CLASS_NAMES.get(class_id, '?')} {confidence*100:.0f}%" # Bounding box color: green for high conf, yellow for moderate box_color = (0, 230, 50) if confidence >= 0.80 else (0, 200, 255) cv2.rectangle(frame, (x, y), (x+w, y+h), box_color, 2) (tw, th), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.55, 1) cv2.rectangle(frame, (x, y - th - 8), (x + tw + 4, y), (0, 0, 0), -1) cv2.putText(frame, label, (x + 2, y - 4), cv2.FONT_HERSHEY_SIMPLEX, 0.55, box_color, 1, cv2.LINE_AA) out.write(frame) cap.release() out.release() return out_path # ───────────────────────────────────────────── # UI HELPERS # ───────────────────────────────────────────── def render_confidence_bars(probs: np.ndarray, top5: np.ndarray): bars_html = '

' for rank, idx in enumerate(top5): p = float(probs[idx]) * 100 name = CLASS_NAMES.get(int(idx), "?") color = "#f5c518" if rank == 0 else ("#888" if p < 5 else "#aaa") width = max(p, 1) bars_html += f"""

{'★ ' if rank==0 else ''}{name} {p:.1f}%

""" bars_html += "

" st.markdown(bars_html, unsafe_allow_html=True) def render_prediction(probs, top5): best_id = int(top5[0]) best_name = CLASS_NAMES.get(best_id, "Unknown") best_conf = float(probs[best_id]) * 100 conf_color = "#4cff91" if best_conf >= 80 else ("#ffb347" if best_conf >= 50 else "#ff6b6b") st.markdown(f"""

Predicted Sign

{best_name}

Confidence: {best_conf:.1f}% · Class ID: {best_id}

""", unsafe_allow_html=True) st.markdown('

Top 5 Predictions

', unsafe_allow_html=True) render_confidence_bars(probs, top5) # ───────────────────────────────────────────── # SIDEBAR # ───────────────────────────────────────────── with st.sidebar: st.markdown("""

🚦 GTSRB

German Traffic Sign Recognition

""", unsafe_allow_html=True) st.markdown('

Model Info

', unsafe_allow_html=True) st.markdown("""

97.8%

Test Acc

Classes

64px

Input Size

CNN

Architecture

""", unsafe_allow_html=True) st.markdown('

Sign Categories

', unsafe_allow_html=True) for cat, ids in CATEGORIES.items(): with st.expander(f"{cat} ({len(ids)} signs)"): for i in ids: st.markdown(f"**{i}** — {CLASS_NAMES[i]}", unsafe_allow_html=True) st.markdown('

Video Detection

', unsafe_allow_html=True) conf_thresh = st.slider("Confidence threshold", 0.30, 0.95, 0.60, 0.05, help="Minimum confidence to draw a bounding box on video frames") st.markdown("""

Dataset: GTSRB (Kaggle)
39,209 training images
Architecture: Custom CNN
Trained locally on GPU

""", unsafe_allow_html=True) # ───────────────────────────────────────────── # MAIN HEADER # ───────────────────────────────────────────── st.markdown("""

German Traffic Sign
Recognition

Deep Learning · 43 Classes · 97.82% Test Accuracy

""", unsafe_allow_html=True) # ───────────────────────────────────────────── # MAIN TABS # ───────────────────────────────────────────── tab_upload, tab_camera, tab_video = st.tabs(["📁 IMAGE UPLOAD", "📷 CAMERA", "🎬 VIDEO DETECTION"]) # ══════════════════════════════════════════════ # TAB 1 — IMAGE UPLOAD # ══════════════════════════════════════════════ with tab_upload: st.markdown('

Upload a Traffic Sign Image

', unsafe_allow_html=True) uploaded = st.file_uploader( "Drag & drop or browse — JPG, JPEG, PNG supported", type=["jpg", "jpeg", "png", "webp"], label_visibility="collapsed" ) if uploaded: pil_img = Image.open(uploaded) col_img, col_pred = st.columns([1, 1.4], gap="large") with col_img: st.markdown('

Uploaded Image

', unsafe_allow_html=True) st.image(pil_img.convert("RGB"), use_container_width=True) st.caption(f"Size: {pil_img.size[0]}×{pil_img.size[1]} px") with col_pred: with st.spinner("Running inference..."): probs, top5 = predict(pil_img) render_prediction(probs, top5) else: st.markdown("""

🚧

Upload an image to begin

""", unsafe_allow_html=True) # ══════════════════════════════════════════════ # TAB 2 — CAMERA INPUT # ══════════════════════════════════════════════ with tab_camera: st.markdown('

Take a Photo

', unsafe_allow_html=True) st.markdown('Point your camera at a German traffic sign and capture', unsafe_allow_html=True) cam_img = st.camera_input("", label_visibility="collapsed") if cam_img: pil_img = Image.open(cam_img) col_c1, col_c2 = st.columns([1, 1.4], gap="large") with col_c1: st.markdown('

Captured Image

', unsafe_allow_html=True) st.image(pil_img.convert("RGB"), use_container_width=True) with col_c2: with st.spinner("Classifying..."): probs, top5 = predict(pil_img) render_prediction(probs, top5) # ══════════════════════════════════════════════ # TAB 3 — VIDEO DETECTION # ══════════════════════════════════════════════ with tab_video: st.markdown('

Upload a Video for Traffic Sign Detection

', unsafe_allow_html=True) st.markdown(""" The system detects red-colored regions in each frame using HSV segmentation, crops candidate regions, classifies them with the CNN, and draws bounding boxes when confidence ≥ threshold set in sidebar. """, unsafe_allow_html=True) vid_file = st.file_uploader( "Upload MP4 / AVI / MOV", type=["mp4", "avi", "mov"], label_visibility="collapsed", key="vid_uploader" ) if vid_file: st.markdown('

Original Video

', unsafe_allow_html=True) st.video(vid_file) if st.button("🔍 Run Detection", use_container_width=True): with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as tmp_in: tmp_in.write(vid_file.read()) tmp_in_path = tmp_in.name with st.spinner("Processing frames... this may take a moment ⏳"): out_path = process_video_frames(tmp_in_path, conf_thresh=conf_thresh) if out_path and os.path.exists(out_path): st.markdown('

Processed Video

', unsafe_allow_html=True) with open(out_path, "rb") as f: st.download_button( "⬇️ Download Processed Video", data=f, file_name="gtsrb_detection_output.mp4", mime="video/mp4", use_container_width=True ) st.video(out_path) else: st.error("Video işlenemedi. Codec veya dosya formatı sorunu olabilir.") else: st.markdown("""

🎬

Upload a video to detect traffic signs

""", unsafe_allow_html=True)