kinemotion 0.10.12__py3-none-any.whl → 0.11.1__py3-none-any.whl

kinemotion/cmj/analysis.py

@@ -0,0 +1,532 @@
+"""Phase detection logic for Counter Movement Jump (CMJ) analysis."""
+
+from enum import Enum
+
+import numpy as np
+from scipy.signal import savgol_filter
+
+from ..core.smoothing import compute_acceleration_from_derivative
+
+
+def compute_signed_velocity(
+    positions: np.ndarray, window_length: int = 5, polyorder: int = 2
+) -> np.ndarray:
+    """
+    Compute SIGNED velocity for CMJ phase detection.
+
+    Unlike drop jump which uses absolute velocity, CMJ needs signed velocity to
+    distinguish upward (negative) from downward (positive) motion.
+
+    Args:
+        positions: 1D array of y-positions in normalized coordinates
+        window_length: Window size for Savitzky-Golay filter
+        polyorder: Polynomial order
+
+    Returns:
+        Signed velocity array where:
+        - Negative = upward motion (y decreasing, jumping up)
+        - Positive = downward motion (y increasing, squatting/falling)
+    """
+    if len(positions) < window_length:
+        return np.diff(positions, prepend=positions[0])
+
+    if window_length % 2 == 0:
+        window_length += 1
+
+    velocity = savgol_filter(
+        positions, window_length, polyorder, deriv=1, delta=1.0, mode="interp"
+    )
+
+    return velocity
+
+
+class CMJPhase(Enum):
+    """Phases of a counter movement jump."""
+
+    STANDING = "standing"
+    ECCENTRIC = "eccentric"  # Downward movement
+    TRANSITION = "transition"  # At lowest point
+    CONCENTRIC = "concentric"  # Upward movement
+    FLIGHT = "flight"
+    LANDING = "landing"
+    UNKNOWN = "unknown"
+
+
+def find_standing_phase(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    fps: float,
+    min_standing_duration: float = 0.5,
+    velocity_threshold: float = 0.01,
+) -> int | None:
+    """
+    Find the end of standing phase (start of countermovement).
+
+    Looks for a period of low velocity (standing) followed by consistent downward motion.
+
+    Args:
+        positions: Array of vertical positions (normalized 0-1)
+        velocities: Array of vertical velocities
+        fps: Video frame rate
+        min_standing_duration: Minimum standing duration in seconds (default: 0.5s)
+        velocity_threshold: Velocity threshold for standing detection
+
+    Returns:
+        Frame index where countermovement begins, or None if not detected.
+    """
+    min_standing_frames = int(fps * min_standing_duration)
+
+    if len(positions) < min_standing_frames:
+        return None
+
+    # Find periods of low velocity (standing)
+    is_standing = np.abs(velocities) < velocity_threshold
+
+    # Look for first sustained standing period
+    standing_count = 0
+    standing_end = None
+
+    for i in range(len(is_standing)):
+        if is_standing[i]:
+            standing_count += 1
+            if standing_count >= min_standing_frames:
+                standing_end = i
+        else:
+            if standing_end is not None:
+                # Found end of standing phase
+                return standing_end
+            standing_count = 0
+
+    return None
+
+
+def find_countermovement_start(
+    velocities: np.ndarray,
+    countermovement_threshold: float = 0.015,
+    min_eccentric_frames: int = 3,
+    standing_start: int | None = None,
+) -> int | None:
+    """
+    Find the start of countermovement (eccentric phase).
+
+    Detects when velocity becomes consistently positive (downward motion in normalized coords).
+
+    Args:
+        velocities: Array of SIGNED vertical velocities
+        countermovement_threshold: Velocity threshold for detecting downward motion (POSITIVE)
+        min_eccentric_frames: Minimum consecutive frames of downward motion
+        standing_start: Optional frame where standing phase ended
+
+    Returns:
+        Frame index where countermovement begins, or None if not detected.
+    """
+    start_frame = standing_start if standing_start is not None else 0
+
+    # Look for sustained downward velocity (POSITIVE in normalized coords)
+    is_downward = velocities[start_frame:] > countermovement_threshold
+    consecutive_count = 0
+
+    for i in range(len(is_downward)):
+        if is_downward[i]:
+            consecutive_count += 1
+            if consecutive_count >= min_eccentric_frames:
+                # Found start of eccentric phase
+                return start_frame + i - consecutive_count + 1
+        else:
+            consecutive_count = 0
+
+    return None
+
+
+def find_lowest_point(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    min_search_frame: int = 80,
+) -> int:
+    """
+    Find the lowest point of countermovement (transition from eccentric to concentric).
+
+    The lowest point occurs BEFORE the peak height (the jump apex). It's where
+    velocity crosses from positive (downward/squatting) to negative (upward/jumping).
+
+    Args:
+        positions: Array of vertical positions (higher value = lower in video)
+        velocities: Array of SIGNED vertical velocities (positive=down, negative=up)
+        min_search_frame: Minimum frame to start searching (default: frame 80)
+
+    Returns:
+        Frame index of lowest point.
+    """
+    # First, find the peak height (minimum y value = highest jump point)
+    peak_height_frame = int(np.argmin(positions))
+
+    # Lowest point MUST be before peak height
+    # Search from min_search_frame to peak_height_frame
+    start_frame = min_search_frame
+    end_frame = peak_height_frame
+
+    if end_frame <= start_frame:
+        start_frame = int(len(positions) * 0.3)
+        end_frame = int(len(positions) * 0.7)
+
+    search_positions = positions[start_frame:end_frame]
+
+    if len(search_positions) == 0:
+        return start_frame
+
+    # Find maximum position value in this range (lowest point in video)
+    lowest_idx = int(np.argmax(search_positions))
+    lowest_frame = start_frame + lowest_idx
+
+    return lowest_frame
+
+
+def refine_transition_with_curvature(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    initial_frame: int,
+    transition_type: str,
+    search_radius: int = 3,
+    window_length: int = 5,
+    polyorder: int = 2,
+) -> float:
+    """
+    Refine transition frame using trajectory curvature (acceleration patterns).
+
+    Uses acceleration (second derivative) to identify characteristic patterns:
+    - Landing: Large acceleration spike (impact deceleration)
+    - Takeoff: Acceleration change (transition from static to flight)
+
+    Args:
+        positions: Array of vertical positions
+        velocities: Array of vertical velocities
+        initial_frame: Initial estimate of transition frame
+        transition_type: Type of transition ("takeoff" or "landing")
+        search_radius: Frames to search around initial estimate (±radius)
+        window_length: Window size for acceleration calculation
+        polyorder: Polynomial order for Savitzky-Golay filter
+
+    Returns:
+        Refined fractional frame index.
+    """
+    # Compute acceleration using second derivative
+    acceleration = compute_acceleration_from_derivative(
+        positions, window_length=window_length, polyorder=polyorder
+    )
+
+    # Define search window
+    search_start = max(0, initial_frame - search_radius)
+    search_end = min(len(positions), initial_frame + search_radius + 1)
+
+    if search_start >= search_end:
+        return float(initial_frame)
+
+    search_accel = acceleration[search_start:search_end]
+
+    if transition_type == "landing":
+        # Landing: Find maximum absolute acceleration (impact)
+        peak_idx = int(np.argmax(np.abs(search_accel)))
+    elif transition_type == "takeoff":
+        # Takeoff: Find maximum acceleration change
+        accel_change = np.abs(np.diff(search_accel))
+        if len(accel_change) > 0:
+            peak_idx = int(np.argmax(accel_change))
+        else:
+            peak_idx = 0
+    else:
+        return float(initial_frame)
+
+    curvature_frame = search_start + peak_idx
+
+    # Blend curvature-based estimate with velocity-based estimate
+    # 70% curvature, 30% velocity
+    blended_frame = 0.7 * curvature_frame + 0.3 * initial_frame
+
+    return float(blended_frame)
+
+
+def interpolate_threshold_crossing(
+    vel_before: float,
+    vel_after: float,
+    velocity_threshold: float,
+) -> float:
+    """
+    Find fractional offset where velocity crosses threshold between two frames.
+
+    Uses linear interpolation assuming velocity changes linearly between frames.
+
+    Args:
+        vel_before: Velocity at frame boundary N (absolute value)
+        vel_after: Velocity at frame boundary N+1 (absolute value)
+        velocity_threshold: Threshold value
+
+    Returns:
+        Fractional offset from frame N (0.0 to 1.0)
+    """
+    # Handle edge cases
+    if abs(vel_after - vel_before) < 1e-9:  # Velocity not changing
+        return 0.5
+
+    # Linear interpolation
+    t = (velocity_threshold - vel_before) / (vel_after - vel_before)
+
+    # Clamp to [0, 1] range
+    return float(max(0.0, min(1.0, t)))
+
+
+def find_cmj_takeoff_from_velocity_peak(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    lowest_point_frame: int,
+    fps: float,
+    window_length: int = 5,
+    polyorder: int = 2,
+) -> float:
+    """
+    Find CMJ takeoff frame as peak upward velocity during concentric phase.
+
+    Takeoff occurs at maximum push-off velocity (most negative velocity),
+    just as feet leave the ground. This is BEFORE peak height is reached.
+
+    Args:
+        positions: Array of vertical positions
+        velocities: Array of SIGNED vertical velocities (negative = upward)
+        lowest_point_frame: Frame at lowest point
+        fps: Video frame rate
+        window_length: Window size for derivative calculations
+        polyorder: Polynomial order for Savitzky-Golay filter
+
+    Returns:
+        Takeoff frame with fractional precision.
+    """
+    concentric_start = int(lowest_point_frame)
+    search_duration = int(
+        fps * 0.3
+    )  # Search next 0.3 seconds (concentric to takeoff is brief)
+    search_end = min(len(velocities), concentric_start + search_duration)
+
+    if search_end <= concentric_start:
+        return float(concentric_start + 1)
+
+    # Find peak upward velocity (most NEGATIVE velocity)
+    # In normalized coords: negative velocity = y decreasing = jumping up
+    concentric_velocities = velocities[concentric_start:search_end]
+    takeoff_idx = int(
+        np.argmin(concentric_velocities)
+    )  # Most negative = fastest upward = takeoff
+    takeoff_frame = concentric_start + takeoff_idx
+
+    return float(takeoff_frame)
+
+
+def find_cmj_landing_from_position_peak(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    accelerations: np.ndarray,
+    takeoff_frame: int,
+    fps: float,
+) -> float:
+    """
+    Find CMJ landing frame by detecting impact after peak height.
+
+    Landing occurs when feet contact ground after peak height, detected by
+    finding where velocity transitions from negative (still going up/at peak)
+    to positive (falling) and position stabilizes.
+
+    Args:
+        positions: Array of vertical positions
+        velocities: Array of SIGNED vertical velocities (negative = up, positive = down)
+        accelerations: Array of accelerations (second derivative)
+        takeoff_frame: Frame at takeoff
+        fps: Video frame rate
+
+    Returns:
+        Landing frame with fractional precision.
+    """
+    # Find peak height (minimum position value in normalized coords)
+    search_start = int(takeoff_frame)
+    search_duration = int(fps * 0.7)  # Search next 0.7 seconds for peak
+    search_end = min(len(positions), search_start + search_duration)
+
+    if search_end <= search_start:
+        return float(search_start + int(fps * 0.3))
+
+    # Find peak height (minimum y value = highest point in frame)
+    flight_positions = positions[search_start:search_end]
+    peak_idx = int(np.argmin(flight_positions))
+    peak_frame = search_start + peak_idx
+
+    # After peak, look for landing (impact with ground)
+    # Landing is detected by maximum positive acceleration (deceleration on impact)
+    landing_search_start = peak_frame + 2
+    landing_search_end = min(len(accelerations), landing_search_start + int(fps * 0.5))
+
+    if landing_search_end <= landing_search_start:
+        return float(peak_frame + int(fps * 0.2))
+
+    # Find impact: maximum positive acceleration after peak
+    # Positive acceleration = slowing down upward motion or impact deceleration
+    landing_accelerations = accelerations[landing_search_start:landing_search_end]
+    impact_idx = int(np.argmax(landing_accelerations))  # Max positive = impact
+    landing_frame = landing_search_start + impact_idx
+
+    return float(landing_frame)
+
+
+def find_interpolated_takeoff_landing(
+    positions: np.ndarray,
+    velocities: np.ndarray,
+    lowest_point_frame: int,
+    window_length: int = 5,
+    polyorder: int = 2,
+) -> tuple[float, float] | None:
+    """
+    Find takeoff and landing frames for CMJ using physics-based detection.
+
+    CMJ-specific: Takeoff is detected as peak velocity (end of push-off),
+    not as high velocity threshold (which detects mid-flight).
+
+    Args:
+        positions: Array of vertical positions
+        velocities: Array of vertical velocities
+        lowest_point_frame: Frame at lowest point
+        window_length: Window size for derivative calculations
+        polyorder: Polynomial order for Savitzky-Golay filter
+
+    Returns:
+        Tuple of (takeoff_frame, landing_frame) with fractional precision, or None.
+    """
+    # Get FPS from velocity array length and assumed duration
+    # This is approximate but sufficient for search windows
+    fps = 30.0  # Default assumption
+
+    # Compute accelerations for landing detection
+    accelerations = compute_acceleration_from_derivative(
+        positions, window_length=window_length, polyorder=polyorder
+    )
+
+    # Find takeoff using peak velocity method (CMJ-specific)
+    takeoff_frame = find_cmj_takeoff_from_velocity_peak(
+        positions, velocities, lowest_point_frame, fps, window_length, polyorder
+    )
+
+    # Find landing using position peak and impact detection
+    landing_frame = find_cmj_landing_from_position_peak(
+        positions, velocities, accelerations, int(takeoff_frame), fps
+    )
+
+    return (takeoff_frame, landing_frame)
+
+
+def _find_takeoff_frame(
+    velocities: np.ndarray, peak_height_frame: int, fps: float
+) -> float:
+    """Find takeoff frame as peak upward velocity before peak height."""
+    takeoff_search_start = max(0, peak_height_frame - int(fps * 0.35))
+    takeoff_search_end = peak_height_frame - 2
+
+    takeoff_velocities = velocities[takeoff_search_start:takeoff_search_end]
+
+    if len(takeoff_velocities) > 0:
+        peak_vel_idx = int(np.argmin(takeoff_velocities))
+        return float(takeoff_search_start + peak_vel_idx)
+    else:
+        return float(peak_height_frame - int(fps * 0.3))
+
+
+def _find_lowest_frame(
+    velocities: np.ndarray, positions: np.ndarray, takeoff_frame: float, fps: float
+) -> float:
+    """Find lowest point frame before takeoff."""
+    lowest_search_start = max(0, int(takeoff_frame) - int(fps * 0.4))
+    lowest_search_end = int(takeoff_frame)
+
+    # Find where velocity crosses from positive to negative
+    for i in range(lowest_search_end - 1, lowest_search_start, -1):
+        if i > 0 and velocities[i] < 0 and velocities[i - 1] >= 0:
+            return float(i)
+
+    # Fallback: use maximum position
+    lowest_positions = positions[lowest_search_start:lowest_search_end]
+    if len(lowest_positions) > 0:
+        lowest_idx = int(np.argmax(lowest_positions))
+        return float(lowest_search_start + lowest_idx)
+    else:
+        return float(int(takeoff_frame) - int(fps * 0.2))
+
+
+def _find_landing_frame(
+    accelerations: np.ndarray, peak_height_frame: int, fps: float
+) -> float:
+    """Find landing frame after peak height."""
+    landing_search_start = peak_height_frame
+    landing_search_end = min(len(accelerations), peak_height_frame + int(fps * 0.5))
+    landing_accelerations = accelerations[landing_search_start:landing_search_end]
+
+    if len(landing_accelerations) > 0:
+        landing_idx = int(np.argmin(landing_accelerations))
+        return float(landing_search_start + landing_idx)
+    else:
+        return float(peak_height_frame + int(fps * 0.3))
+
+
+def _find_standing_end(velocities: np.ndarray, lowest_point: float) -> float | None:
+    """Find end of standing phase before lowest point."""
+    if lowest_point <= 20:
+        return None
+
+    standing_search = velocities[: int(lowest_point)]
+    low_vel = np.abs(standing_search) < 0.005
+    if np.any(low_vel):
+        standing_frames = np.nonzero(low_vel)[0]
+        if len(standing_frames) > 10:
+            return float(standing_frames[-1])
+
+    return None
+
+
+def detect_cmj_phases(
+    positions: np.ndarray,
+    fps: float,
+    window_length: int = 5,
+    polyorder: int = 2,
+) -> tuple[float | None, float, float, float] | None:
+    """
+    Detect all phases of a counter movement jump using a simplified, robust approach.
+
+    Strategy: Work BACKWARD from peak height to find all phases.
+    1. Find peak height (global minimum y)
+    2. Find takeoff (peak negative velocity before peak height)
+    3. Find lowest point (maximum y value before takeoff)
+    4. Find landing (impact after peak height)
+
+    Args:
+        positions: Array of vertical positions (normalized 0-1)
+        fps: Video frame rate
+        window_length: Window size for derivative calculations
+        polyorder: Polynomial order for Savitzky-Golay filter
+
+    Returns:
+        Tuple of (standing_end_frame, lowest_point_frame, takeoff_frame, landing_frame)
+        with fractional precision, or None if phases cannot be detected.
+    """
+    # Compute SIGNED velocities and accelerations
+    velocities = compute_signed_velocity(
+        positions, window_length=window_length, polyorder=polyorder
+    )
+    accelerations = compute_acceleration_from_derivative(
+        positions, window_length=window_length, polyorder=polyorder
+    )
+
+    # Step 1: Find peak height (global minimum y = highest point in frame)
+    peak_height_frame = int(np.argmin(positions))
+    if peak_height_frame < 10:
+        return None  # Peak too early, invalid
+
+    # Step 2-4: Find all phases using helper functions
+    takeoff_frame = _find_takeoff_frame(velocities, peak_height_frame, fps)
+    lowest_point = _find_lowest_frame(velocities, positions, takeoff_frame, fps)
+    landing_frame = _find_landing_frame(accelerations, peak_height_frame, fps)
+    standing_end = _find_standing_end(velocities, lowest_point)
+
+    return (standing_end, lowest_point, takeoff_frame, landing_frame)