import cv2 import mediapipe as mp import numpy as np import math, time, json from pythonosc import udp_client # ------------------------------- # SETTINGS # ------------------------------- TOUCH_CAM_INDEX = 0 GESTURE_CAM_INDEX = 1 GAME_SCREEN_WIDTH = 900 GAME_SCREEN_HEIGHT = 600 STILL_REQUIRED = 1.0 MOVE_TOLERANCE = 25 TARGET_FPS = 20 FRAME_TIME = 1.0 / TARGET_FPS client = udp_client.SimpleUDPClient("127.0.0.1", 5005) # Global states last_finger_pos = None finger_still_start = None prev_touch_time = 0.0 prev_clap_time = 0.0 # ------------------------------------- # LOAD CALIBRATION + HOMOGRAPHY # ------------------------------------- try: with open("calibration.json", "r") as f: CALIB_POINTS = json.load(f) print("📐 Kalibrierung geladen:", CALIB_POINTS) except: CALIB_POINTS = None print("⚠️ Keine Kalibrierung gefunden – benutze Rohkoordinaten!") H = None if CALIB_POINTS is not None: src = np.array(CALIB_POINTS, dtype=np.float32) dst = np.array([ [0, 0], [GAME_SCREEN_WIDTH, 0], [GAME_SCREEN_WIDTH, GAME_SCREEN_HEIGHT], [0, GAME_SCREEN_HEIGHT] ], dtype=np.float32) H, _ = cv2.findHomography(src, dst) print("📐 Homographie-Matrix berechnet!") def map_point_homography(x, y): if H is None: return int(x), int(y) p = np.array([[[x, y]]], dtype=np.float32) mapped = cv2.perspectiveTransform(p, H)[0][0] return int(mapped[0]), int(mapped[1]) # ----------------------------------------------------------------- def run_gesture_input(): global last_finger_pos, finger_still_start global prev_touch_time, prev_clap_time mp_hands = mp.solutions.hands mp_draw = mp.solutions.drawing_utils # FASTER HAND MODELS hands_touch = mp_hands.Hands(max_num_hands=1, min_detection_confidence=0.6, model_complexity=0) hands_gesture = mp_hands.Hands(max_num_hands=2, min_detection_confidence=0.6, model_complexity=0) # Cameras cam_touch = cv2.VideoCapture(TOUCH_CAM_INDEX) cam_gesture = cv2.VideoCapture(GESTURE_CAM_INDEX) # Set to 640x480 for cam in (cam_touch, cam_gesture): cam.set(cv2.CAP_PROP_FRAME_WIDTH, 640) cam.set(cv2.CAP_PROP_FRAME_HEIGHT, 480) if not cam_touch.isOpened(): print("❌ Touch-Kamera konnte NICHT geöffnet werden!") if not cam_gesture.isOpened(): print("❌ Gesture-Kamera konnte NICICHT geöffnet werden!") clap_cooldown = 1.5 while True: loop_start = time.time() ok1, frame_touch = cam_touch.read() ok2, frame_gest = cam_gesture.read() if not ok1 or not ok2: print("❌ Eine Kamera liefert kein Bild.") break # Flip für Orientierung frame_touch = cv2.flip(frame_touch, -1) frame_gest = cv2.flip(frame_gest, 1) # ---------------- TOUCH detection ---------------- rgb_t = cv2.cvtColor(frame_touch, cv2.COLOR_BGR2RGB) res_t = hands_touch.process(rgb_t) th, tw, _ = frame_touch.shape if res_t.multi_hand_landmarks: lm = res_t.multi_hand_landmarks[0] # Finger zeigt nach unten? if lm.landmark[8].y < lm.landmark[5].y: last_finger_pos = None finger_still_start = None continue fx = int(lm.landmark[8].x * tw) fy = int(lm.landmark[8].y * th) sx, sy = map_point_homography(fx, fy) now = time.time() current_pos = (fx, fy) if last_finger_pos is None: last_finger_pos = current_pos finger_still_start = now else: dist = math.hypot(current_pos[0] - last_finger_pos[0], current_pos[1] - last_finger_pos[1]) if dist < MOVE_TOLERANCE: if finger_still_start is None: finger_still_start = now else: still_time = now - finger_still_start if still_time >= STILL_REQUIRED and (now - prev_touch_time) > 0.5: client.send_message("/touch", [sx, sy]) print(f"👉 TOUCH bei {sx},{sy} nach {still_time:.2f}s") prev_touch_time = now finger_still_start = None else: finger_still_start = now last_finger_pos = current_pos cv2.circle(frame_touch, (fx, fy), 10, (0, 255, 0), -1) cv2.putText(frame_touch, f"{sx},{sy}", (fx + 10, fy - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,255,0), 2) else: last_finger_pos = None finger_still_start = None # ---------------- Clap detection ---------------- rgb_g = cv2.cvtColor(frame_gest, cv2.COLOR_BGR2RGB) res_g = hands_gesture.process(rgb_g) gh, gw, _ = frame_gest.shape if res_g.multi_hand_landmarks and len(res_g.multi_hand_landmarks) == 2: h1, h2 = res_g.multi_hand_landmarks x1 = np.mean([p.x for p in h1.landmark]) * gw y1 = np.mean([p.y for p in h1.landmark]) * gh x2 = np.mean([p.x for p in h2.landmark]) * gw y2 = np.mean([p.y for p in h2.landmark]) * gh dist = math.hypot(x2 - x1, y2 - y1) if dist < 100 and (time.time() - prev_clap_time) > clap_cooldown: prev_clap_time = time.time() client.send_message("/clap", 1) print("👏 SEND /clap") cv2.putText(frame_gest, "👏", (int(gw/2)-20, 80), cv2.FONT_HERSHEY_SIMPLEX, 2, (0,255,255), 3) # Display cv2.imshow("Touch-Cam", frame_touch) cv2.imshow("Gesture-Cam", frame_gest) # -------------- FPS LIMITER -------------- elapsed = time.time() - loop_start sleep_time = FRAME_TIME - elapsed if sleep_time > 0: time.sleep(sleep_time) if cv2.waitKey(1) & 0xFF == 27: break cam_touch.release() cam_gesture.release() cv2.destroyAllWindows() if __name__ == "__main__": run_gesture_input()