global_match_memory/gesture_input_osc_test3.py
2025-12-11 11:22:35 +01:00

197 lines
6.0 KiB
Python

import cv2
import mediapipe as mp
import numpy as np
import math, time, json
from pythonosc import udp_client
# =====================================================
# =================== SETTINGS ========================
# =====================================================
# -------- Camera Index --------
TOUCH_CAM_INDEX = 1
GESTURE_CAM_INDEX = 0
# -------- Camera Capture Resolution / FPS --------
CAM_WIDTH = 1280
CAM_HEIGHT = 720
CAM_FPS = 30
# -------- Display Resolution (INTEGER) --------
DISPLAY_WIDTH = 480 #960
DISPLAY_HEIGHT = 270 #540
# -------- Screen Mapping --------
GAME_SCREEN_WIDTH = 900
GAME_SCREEN_HEIGHT = 600
# -------- MediaPipe Model Complexity --------
MODEL_COMPLEXITY_TOUCH = 1
MODEL_COMPLEXITY_GESTURE = 0
# -------- Touch Trigger --------
STILL_REQUIRED = 1.0
MOVE_TOLERANCE = 25
TOUCH_COOLDOWN = 0.5
# -------- Clap Trigger --------
CLAP_DISTANCE = 100
CLAP_COOLDOWN = 1
# -------- OSC --------
OSC_IP = "127.0.0.1"
OSC_PORT = 5005
# =====================================================
# ================= GLOBAL STATE ======================
# =====================================================
client = udp_client.SimpleUDPClient(OSC_IP, OSC_PORT)
last_finger_pos = None
finger_still_start = None
prev_touch_time = 0.0
prev_clap_time = 0.0
# =====================================================
# ============ CALIBRATION / HOMOGRAPHY ===============
# =====================================================
try:
with open("calibration.json", "r") as f:
CALIB_POINTS = json.load(f)
print("📐 Calibration loaded")
except:
CALIB_POINTS = None
print("⚠️ No calibration found")
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)
def map_point_homography(x, y):
if H is None:
return int(x), int(y)
p = np.array([[[x, y]]], dtype=np.float32)
m = cv2.perspectiveTransform(p, H)[0][0]
return int(m[0]), int(m[1])
# =====================================================
# ===================== MAIN ==========================
# =====================================================
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
hands_touch = mp_hands.Hands(
max_num_hands=1,
model_complexity=MODEL_COMPLEXITY_TOUCH,
min_detection_confidence=0.6,
min_tracking_confidence=0.6
)
hands_gesture = mp_hands.Hands(
max_num_hands=2,
model_complexity=MODEL_COMPLEXITY_GESTURE,
min_detection_confidence=0.6,
min_tracking_confidence=0.6
)
cam_touch = cv2.VideoCapture(TOUCH_CAM_INDEX)
cam_gesture = cv2.VideoCapture(GESTURE_CAM_INDEX)
for cam in (cam_touch, cam_gesture):
cam.set(cv2.CAP_PROP_FRAME_WIDTH, CAM_WIDTH)
cam.set(cv2.CAP_PROP_FRAME_HEIGHT, CAM_HEIGHT)
cam.set(cv2.CAP_PROP_FPS, CAM_FPS)
while True:
ok1, frame_touch = cam_touch.read()
ok2, frame_gest = cam_gesture.read()
if not ok1 or not ok2:
break
frame_touch = cv2.flip(frame_touch, -1)
frame_gest = cv2.flip(frame_gest, 1)
# ---------------- TOUCH ----------------
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]
mp_draw.draw_landmarks(frame_touch, lm, mp_hands.HAND_CONNECTIONS)
if lm.landmark[8].y >= lm.landmark[5].y:
fx = int(lm.landmark[8].x * tw)
fy = int(lm.landmark[8].y * th)
sx, sy = map_point_homography(fx, fy)
now = time.time()
cur = (fx, fy)
if last_finger_pos is None:
last_finger_pos = cur
finger_still_start = now
else:
dist = math.hypot(cur[0]-last_finger_pos[0], cur[1]-last_finger_pos[1])
if dist < MOVE_TOLERANCE:
if now - finger_still_start >= STILL_REQUIRED and now - prev_touch_time > TOUCH_COOLDOWN:
client.send_message("/touch", [sx, sy])
prev_touch_time = now
finger_still_start = now
else:
finger_still_start = now
last_finger_pos = cur
cv2.circle(frame_touch, (fx, fy), 10, (0,255,0), -1)
else:
last_finger_pos = None
# ---------------- CLAP ----------------
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
c1 = np.mean([[p.x*gw, p.y*gh] for p in h1.landmark], axis=0)
c2 = np.mean([[p.x*gw, p.y*gh] for p in h2.landmark], axis=0)
dist = np.linalg.norm(c2 - c1)
if dist < CLAP_DISTANCE and time.time() - prev_clap_time > CLAP_COOLDOWN:
prev_clap_time = time.time()
client.send_message("/clap", 1)
# ---------------- DISPLAY SCALING ----------------
disp_touch = cv2.resize(frame_touch, (DISPLAY_WIDTH, DISPLAY_HEIGHT))
disp_gest = cv2.resize(frame_gest, (DISPLAY_WIDTH, DISPLAY_HEIGHT))
cv2.imshow("Touch Camera", disp_touch)
cv2.imshow("Gesture Camera", disp_gest)
if cv2.waitKey(5) & 0xFF == 27:
break
cam_touch.release()
cam_gesture.release()
cv2.destroyAllWindows()
# =====================================================
if __name__ == "__main__":
run_gesture_input()