mirror of
https://github.com/kavishdevar/librepods.git
synced 2026-01-29 06:10:52 +00:00
358 lines
16 KiB
Python
358 lines
16 KiB
Python
import logging
|
|
import statistics
|
|
import time
|
|
from bluetooth import BluetoothSocket
|
|
from collections import deque
|
|
from colors import *
|
|
from connection_manager import ConnectionManager
|
|
from logging import Logger, StreamHandler
|
|
from threading import Lock, Thread
|
|
from typing import Any, Deque, List, Optional, Tuple
|
|
|
|
handler: StreamHandler = StreamHandler()
|
|
handler.setFormatter(ColorFormatter())
|
|
log: Logger = logging.getLogger(__name__)
|
|
log.setLevel(logging.INFO)
|
|
log.addHandler(handler)
|
|
log.propagate = False
|
|
|
|
class GestureDetector:
|
|
INIT_CMD: str = "00 00 04 00 01 00 02 00 00 00 00 00 00 00 00 00"
|
|
START_CMD: str = "04 00 04 00 17 00 00 00 10 00 10 00 08 A1 02 42 0B 08 0E 10 02 1A 05 01 40 9C 00 00"
|
|
STOP_CMD: str = "04 00 04 00 17 00 00 00 10 00 11 00 08 7E 10 02 42 0B 08 4E 10 02 1A 05 01 00 00 00 00"
|
|
|
|
def __init__(self, conn: ConnectionManager = None) -> None:
|
|
self.sock: BluetoothSocket = None
|
|
self.bt_addr: str = "28:2D:7F:C2:05:5B"
|
|
self.psm: int = 0x1001
|
|
self.running: bool = False
|
|
self.data_lock: Lock = Lock()
|
|
|
|
self.horiz_buffer: Deque[int] = deque(maxlen=100)
|
|
self.vert_buffer: Deque[int] = deque(maxlen=100)
|
|
|
|
self.horiz_avg_buffer: Deque[float] = deque(maxlen=5)
|
|
self.vert_avg_buffer: Deque[float] = deque(maxlen=5)
|
|
|
|
self.horiz_peaks: List[int] = []
|
|
self.horiz_troughs: List[int] = []
|
|
self.vert_peaks: List[int] = []
|
|
self.vert_troughs: List[int] = []
|
|
|
|
self.last_peak_time: float = 0
|
|
self.peak_intervals: Deque[float] = deque(maxlen=5)
|
|
|
|
self.peak_threshold: int = 400
|
|
self.direction_change_threshold: int = 175
|
|
self.rhythm_consistency_threshold: float = 0.5
|
|
|
|
self.horiz_increasing: Optional[bool] = None
|
|
self.vert_increasing: Optional[bool] = None
|
|
|
|
self.required_extremes = 3
|
|
self.detection_timeout: int = 15
|
|
|
|
self.min_confidence_threshold: float = 0.7
|
|
|
|
self.conn: ConnectionManager = conn
|
|
|
|
def connect(self) -> bool:
|
|
try:
|
|
log.info(f"Connecting to AirPods at {self.bt_addr}...")
|
|
if self.conn is None:
|
|
self.conn = ConnectionManager(self.bt_addr, self.psm, logger=log)
|
|
if not self.conn.connect():
|
|
return False
|
|
else:
|
|
if not self.conn.connected:
|
|
if not self.conn.connect():
|
|
return False
|
|
self.sock = self.conn.sock
|
|
log.info(f"{Colors.GREEN}✓ Connected to AirPods via ConnectionManager{Colors.RESET}")
|
|
return True
|
|
except Exception as e:
|
|
log.error(f"{Colors.RED}Connection failed: {e}{Colors.RESET}")
|
|
return False
|
|
|
|
def process_data(self) -> None:
|
|
"""Process incoming head tracking data."""
|
|
self.conn.send_start()
|
|
log.info(f"{Colors.GREEN}✓ Head tracking activated{Colors.RESET}")
|
|
|
|
self.running = True
|
|
start_time: float = time.time()
|
|
|
|
log.info(f"{Colors.GREEN}Ready! Make a YES or NO gesture{Colors.RESET}")
|
|
log.info(f"{Colors.YELLOW}Tip: Use natural, moderate speed head movements{Colors.RESET}")
|
|
|
|
while self.running:
|
|
if time.time() - start_time > self.detection_timeout:
|
|
log.warning(f"{Colors.YELLOW}⚠️ Detection timeout reached. No gesture detected.{Colors.RESET}")
|
|
self.running = False
|
|
break
|
|
|
|
try:
|
|
if not self.sock:
|
|
log.error("Socket not available.")
|
|
break
|
|
data: bytes = self.sock.recv(1024)
|
|
formatted: str = self.format_hex(data)
|
|
if self.is_valid_tracking_packet(formatted):
|
|
raw_bytes: bytes = bytes.fromhex(formatted.replace(" ", ""))
|
|
horizontal, vertical = self.extract_orientation_values(raw_bytes)
|
|
|
|
if horizontal is not None and vertical is not None:
|
|
smooth_h, smooth_v = self.apply_smoothing(horizontal, vertical)
|
|
|
|
with self.data_lock:
|
|
self.horiz_buffer.append(smooth_h)
|
|
self.vert_buffer.append(smooth_v)
|
|
|
|
self.detect_peaks_and_troughs()
|
|
gesture: Optional[str] = self.detect_gestures()
|
|
|
|
if gesture:
|
|
self.running = False
|
|
break
|
|
|
|
except Exception as e:
|
|
if self.running:
|
|
log.error(f"Data processing error: {e}")
|
|
break
|
|
|
|
def disconnect(self) -> None:
|
|
"""Disconnect from socket."""
|
|
self.conn.disconnect()
|
|
|
|
def format_hex(self, data: bytes) -> str:
|
|
"""Format binary data to readable hex string."""
|
|
hex_str: str = data.hex()
|
|
return ' '.join(hex_str[i:i+2] for i in range(0, len(hex_str), 2))
|
|
|
|
def is_valid_tracking_packet(self, hex_string: str) -> bool:
|
|
"""Verify packet is a valid head tracking packet."""
|
|
standard_header: str = "04 00 04 00 17 00 00 00 10 00 45 00"
|
|
alternate_header: str = "04 00 04 00 17 00 00 00 10 00 44 00"
|
|
if not hex_string.startswith(standard_header) and not hex_string.startswith(alternate_header):
|
|
return False
|
|
|
|
if len(hex_string.split()) < 80:
|
|
return False
|
|
|
|
return True
|
|
|
|
def extract_orientation_values(self, raw_bytes: bytes) -> Tuple[Optional[int], Optional[int]]:
|
|
"""Extract head orientation data from packet."""
|
|
try:
|
|
horizontal: int = int.from_bytes(raw_bytes[51:53], byteorder='little', signed=True)
|
|
vertical: int = int.from_bytes(raw_bytes[53:55], byteorder='little', signed=True)
|
|
|
|
return horizontal, vertical
|
|
except Exception as e:
|
|
log.debug(f"Failed to extract orientation: {e}")
|
|
return None, None
|
|
|
|
def apply_smoothing(self, horizontal: int, vertical: int) -> Tuple[float, float]:
|
|
"""Apply moving average smoothing (Apple-like filtering)."""
|
|
self.horiz_avg_buffer.append(horizontal)
|
|
self.vert_avg_buffer.append(vertical)
|
|
|
|
smooth_horiz: float = sum(self.horiz_avg_buffer) / len(self.horiz_avg_buffer)
|
|
smooth_vert: float = sum(self.vert_avg_buffer) / len(self.vert_avg_buffer)
|
|
|
|
return smooth_horiz, smooth_vert
|
|
|
|
def detect_peaks_and_troughs(self) -> None:
|
|
"""Detect motion direction changes with Apple-like refinements."""
|
|
if len(self.horiz_buffer) < 4 or len(self.vert_buffer) < 4:
|
|
return
|
|
|
|
h_values: List[int] = list(self.horiz_buffer)[-4:]
|
|
v_values: List[int] = list(self.vert_buffer)[-4:]
|
|
|
|
h_variance: float = statistics.variance(h_values) if len(h_values) > 1 else 0
|
|
v_variance: float = statistics.variance(v_values) if len(v_values) > 1 else 0
|
|
|
|
current: int = self.horiz_buffer[-1]
|
|
prev: int = self.horiz_buffer[-2]
|
|
|
|
if self.horiz_increasing is None:
|
|
self.horiz_increasing = current > prev
|
|
|
|
dynamic_h_threshold: float = max(100, min(self.direction_change_threshold, h_variance / 3))
|
|
|
|
if self.horiz_increasing and current < prev - dynamic_h_threshold:
|
|
if abs(prev) > self.peak_threshold:
|
|
self.horiz_peaks.append((len(self.horiz_buffer)-1, prev, time.time()))
|
|
direction: str = "➡️ " if prev > 0 else "⬅️ "
|
|
log.info(f"{Colors.CYAN}{direction} Horizontal max: {prev} (threshold: {dynamic_h_threshold:.1f}){Colors.RESET}")
|
|
|
|
now: float = time.time()
|
|
if self.last_peak_time > 0:
|
|
interval: float = now - self.last_peak_time
|
|
self.peak_intervals.append(interval)
|
|
self.last_peak_time = now
|
|
|
|
self.horiz_increasing = False
|
|
|
|
elif not self.horiz_increasing and current > prev + dynamic_h_threshold:
|
|
if abs(prev) > self.peak_threshold:
|
|
self.horiz_troughs.append((len(self.horiz_buffer)-1, prev, time.time()))
|
|
direction: str = "➡️ " if prev > 0 else "⬅️ "
|
|
log.info(f"{Colors.CYAN}{direction} Horizontal max: {prev} (threshold: {dynamic_h_threshold:.1f}){Colors.RESET}")
|
|
|
|
now: float = time.time()
|
|
if self.last_peak_time > 0:
|
|
interval: float = now - self.last_peak_time
|
|
self.peak_intervals.append(interval)
|
|
self.last_peak_time = now
|
|
|
|
self.horiz_increasing = True
|
|
|
|
current: int = self.vert_buffer[-1]
|
|
prev: int = self.vert_buffer[-2]
|
|
|
|
if self.vert_increasing is None:
|
|
self.vert_increasing = current > prev
|
|
|
|
dynamic_v_threshold: float = max(100, min(self.direction_change_threshold, v_variance / 3))
|
|
|
|
if self.vert_increasing and current < prev - dynamic_v_threshold:
|
|
if abs(prev) > self.peak_threshold:
|
|
self.vert_peaks.append((len(self.vert_buffer)-1, prev, time.time()))
|
|
direction: str = "⬆️ " if prev > 0 else "⬇️ "
|
|
log.info(f"{Colors.MAGENTA}{direction} Vertical max: {prev} (threshold: {dynamic_v_threshold:.1f}){Colors.RESET}")
|
|
|
|
now: float = time.time()
|
|
if self.last_peak_time > 0:
|
|
interval: float = now - self.last_peak_time
|
|
self.peak_intervals.append(interval)
|
|
self.last_peak_time = now
|
|
|
|
self.vert_increasing = False
|
|
|
|
elif not self.vert_increasing and current > prev + dynamic_v_threshold:
|
|
if abs(prev) > self.peak_threshold:
|
|
self.vert_troughs.append((len(self.vert_buffer)-1, prev, time.time()))
|
|
direction: str = "⬆️ " if prev > 0 else "⬇️ "
|
|
log.info(f"{Colors.MAGENTA}{direction} Vertical max: {prev} (threshold: {dynamic_v_threshold:.1f}){Colors.RESET}")
|
|
|
|
now: float = time.time()
|
|
if self.last_peak_time > 0:
|
|
interval: float = now - self.last_peak_time
|
|
self.peak_intervals.append(interval)
|
|
self.last_peak_time = now
|
|
|
|
self.vert_increasing = True
|
|
|
|
def calculate_rhythm_consistency(self) -> float:
|
|
"""Calculate how consistent the timing between peaks is (Apple-like)."""
|
|
if len(self.peak_intervals) < 2:
|
|
return 0
|
|
|
|
mean_interval: float = statistics.mean(self.peak_intervals)
|
|
if mean_interval == 0:
|
|
return 0
|
|
|
|
variances: List[float] = [(i/mean_interval - 1.0) ** 2 for i in self.peak_intervals]
|
|
consistency: float = 1.0 - min(1.0, statistics.mean(variances) / self.rhythm_consistency_threshold)
|
|
return max(0, consistency)
|
|
|
|
def calculate_confidence_score(self, extremes: List[Tuple[int, int, float]], is_vertical: bool = True) -> float:
|
|
"""Calculate confidence score for gesture detection (Apple-like)."""
|
|
if len(extremes) < self.required_extremes:
|
|
return 0.0
|
|
|
|
sorted_extremes: List[Tuple[int, int, float]] = sorted(extremes, key=lambda x: x[0])
|
|
|
|
recent: List[Tuple[int, int, float]] = sorted_extremes[-self.required_extremes:]
|
|
|
|
avg_amplitude: float = sum(abs(val) for _, val, _ in recent) / len(recent)
|
|
amplitude_factor: float = min(1.0, avg_amplitude / 600)
|
|
|
|
rhythm_factor: float = self.calculate_rhythm_consistency()
|
|
|
|
signs: List[int] = [1 if val > 0 else -1 for _, val, _ in recent]
|
|
alternating: bool = all(signs[i] != signs[i-1] for i in range(1, len(signs)))
|
|
alternation_factor: float = 1.0 if alternating else 0.5
|
|
|
|
if is_vertical:
|
|
vert_amp: float = sum(abs(val) for _, val, _ in recent) / len(recent)
|
|
horiz_vals: List[int] = list(self.horiz_buffer)[-len(recent)*2:]
|
|
horiz_amp: float = sum(abs(val) for val in horiz_vals) / len(horiz_vals) if horiz_vals else 0
|
|
isolation_factor: float = min(1.0, vert_amp / (horiz_amp + 0.1) * 1.2)
|
|
else:
|
|
horiz_amp: float = sum(abs(val) for _, val, _ in recent)
|
|
vert_vals: List[int] = list(self.vert_buffer)[-len(recent)*2:]
|
|
vert_amp: float = sum(abs(val) for val in vert_vals) / len(vert_vals) if vert_vals else 0
|
|
isolation_factor: float = min(1.0, horiz_amp / (vert_amp + 0.1) * 1.2)
|
|
|
|
confidence: float = (
|
|
amplitude_factor * 0.4 +
|
|
rhythm_factor * 0.2 +
|
|
alternation_factor * 0.2 +
|
|
isolation_factor * 0.2
|
|
)
|
|
|
|
return confidence
|
|
|
|
def detect_gestures(self) -> Optional[str]:
|
|
"""Recognize head gesture patterns with Apple-like intelligence."""
|
|
if len(self.vert_peaks) + len(self.vert_troughs) >= self.required_extremes:
|
|
all_extremes: List[Tuple[int, int, float]] = sorted(self.vert_peaks + self.vert_troughs, key=lambda x: x[0])
|
|
|
|
confidence: float = self.calculate_confidence_score(all_extremes, is_vertical=True)
|
|
|
|
log.info(f"Vertical motion confidence: {confidence:.2f} (need {self.min_confidence_threshold:.2f})")
|
|
|
|
if confidence >= self.min_confidence_threshold:
|
|
log.info(f"{Colors.GREEN}🎯 \"Yes\" Gesture Detected (confidence: {confidence:.2f}){Colors.RESET}")
|
|
return "YES"
|
|
|
|
if len(self.horiz_peaks) + len(self.horiz_troughs) >= self.required_extremes:
|
|
all_extremes: List[Tuple[int, int, float]] = sorted(self.horiz_peaks + self.horiz_troughs, key=lambda x: x[0])
|
|
|
|
confidence: float = self.calculate_confidence_score(all_extremes, is_vertical=False)
|
|
|
|
log.info(f"Horizontal motion confidence: {confidence:.2f} (need {self.min_confidence_threshold:.2f})")
|
|
|
|
if confidence >= self.min_confidence_threshold:
|
|
log.info(f"{Colors.GREEN}🎯 \"No\" gesture detected (confidence: {confidence:.2f}){Colors.RESET}")
|
|
return "NO"
|
|
|
|
return None
|
|
|
|
def start_detection(self) -> None:
|
|
"""Begin gesture detection process."""
|
|
log.info(f"{Colors.BOLD}{Colors.WHITE}Starting gesture detection...{Colors.RESET}")
|
|
|
|
if not self.connect():
|
|
log.error(f"{Colors.RED}Failed to connect to AirPods.{Colors.RESET}")
|
|
return
|
|
|
|
data_thread: Thread = Thread(target=self.process_data)
|
|
data_thread.daemon = True
|
|
data_thread.start()
|
|
|
|
try:
|
|
data_thread.join(timeout=self.detection_timeout + 2)
|
|
if data_thread.is_alive():
|
|
log.warning(f"{Colors.YELLOW}⚠️ Timeout reached. Stopping detection.{Colors.RESET}")
|
|
self.running = False
|
|
except KeyboardInterrupt:
|
|
log.info(f"{Colors.YELLOW}Detection canceled by user.{Colors.RESET}")
|
|
self.running = False
|
|
if __name__ == "__main__":
|
|
self.disconnect()
|
|
log.info(f"{Colors.GREEN}Gesture detection complete.{Colors.RESET}")
|
|
|
|
if __name__ == "__main__":
|
|
print(f"{Colors.BG_BLACK}{Colors.CYAN}╔════════════════════════════════════════╗{Colors.RESET}")
|
|
print(f"{Colors.BG_BLACK}{Colors.CYAN}║ AirPods Head Gesture Detector ║{Colors.RESET}")
|
|
print(f"{Colors.BG_BLACK}{Colors.CYAN}╚════════════════════════════════════════╝{Colors.RESET}")
|
|
print(f"\n{Colors.WHITE}This program detects head gestures using AirPods:{Colors.RESET}")
|
|
print(f"{Colors.GREEN}• YES: {Colors.WHITE}nodding head up and down{Colors.RESET}")
|
|
print(f"{Colors.RED}• NO: {Colors.WHITE}shaking head left and right{Colors.RESET}\n")
|
|
|
|
detector: GestureDetector = GestureDetector()
|
|
detector.start_detection() |