gaitsetpy.features.urfall_features
UrFall Image/Video Feature Extractor. Maintainer: @aharshit123456
This module provides a lightweight feature extractor for UrFall depth (PNG16), RGB (PNG), and video (MP4) modalities. It extracts simple, fast-to-compute features suitable for sanity checks and baseline models.
1''' 2UrFall Image/Video Feature Extractor. 3Maintainer: @aharshit123456 4 5This module provides a lightweight feature extractor for UrFall depth (PNG16), RGB (PNG), and video (MP4) modalities. 6It extracts simple, fast-to-compute features suitable for sanity checks and baseline models. 7''' 8 9from typing import List, Dict, Any 10import numpy as np 11from ..core.base_classes import BaseFeatureExtractor 12 13 14class UrFallMediaFeatureExtractor(BaseFeatureExtractor): 15 """ 16 UrFall image/video feature extractor. 17 18 Extracts per-window features from sequences of images or decoded video frames: 19 - mean_intensity 20 - std_intensity 21 - motion_mean (if pairwise differences are available) 22 - motion_std 23 """ 24 25 def __init__(self, verbose: bool = False): 26 super().__init__( 27 name="urfall_media", 28 description="Lightweight feature extractor for UrFall depth/RGB/video modalities" 29 ) 30 self.config = { 31 'verbose': verbose, 32 'use_motion': True, 33 'grayscale': True, # for RGB convert to gray before stats 34 } 35 36 def extract_features(self, windows: List[Dict], fs: int, **kwargs) -> List[Dict]: 37 self.config.update(kwargs) 38 features: List[Dict[str, Any]] = [] 39 for window in windows: 40 name = window.get('name', 'unknown') 41 frames = window.get('data', []) 42 if not isinstance(frames, list) or len(frames) == 0: 43 continue 44 # frames is a list of numpy arrays (HxW) or (HxWxC) 45 intensities = [] 46 motions = [] 47 prev_gray = None 48 for f in frames: 49 arr = np.array(f) 50 if arr.ndim == 3 and self.config['grayscale']: 51 arr = arr.mean(axis=2) 52 intensities.append(float(np.mean(arr))) 53 if self.config['use_motion']: 54 if prev_gray is not None: 55 diff = np.abs(arr.astype(np.float32) - prev_gray.astype(np.float32)) 56 motions.append(float(np.mean(diff))) 57 prev_gray = arr 58 fdict: Dict[str, Any] = {'name': name, 'features': {}} 59 if intensities: 60 fdict['features']['mean_intensity'] = float(np.mean(intensities)) 61 fdict['features']['std_intensity'] = float(np.std(intensities)) 62 if motions: 63 fdict['features']['motion_mean'] = float(np.mean(motions)) 64 fdict['features']['motion_std'] = float(np.std(motions)) 65 features.append(fdict) 66 return features 67 68 def get_feature_names(self) -> List[str]: 69 return ['mean_intensity', 'std_intensity', 'motion_mean', 'motion_std']
15class UrFallMediaFeatureExtractor(BaseFeatureExtractor): 16 """ 17 UrFall image/video feature extractor. 18 19 Extracts per-window features from sequences of images or decoded video frames: 20 - mean_intensity 21 - std_intensity 22 - motion_mean (if pairwise differences are available) 23 - motion_std 24 """ 25 26 def __init__(self, verbose: bool = False): 27 super().__init__( 28 name="urfall_media", 29 description="Lightweight feature extractor for UrFall depth/RGB/video modalities" 30 ) 31 self.config = { 32 'verbose': verbose, 33 'use_motion': True, 34 'grayscale': True, # for RGB convert to gray before stats 35 } 36 37 def extract_features(self, windows: List[Dict], fs: int, **kwargs) -> List[Dict]: 38 self.config.update(kwargs) 39 features: List[Dict[str, Any]] = [] 40 for window in windows: 41 name = window.get('name', 'unknown') 42 frames = window.get('data', []) 43 if not isinstance(frames, list) or len(frames) == 0: 44 continue 45 # frames is a list of numpy arrays (HxW) or (HxWxC) 46 intensities = [] 47 motions = [] 48 prev_gray = None 49 for f in frames: 50 arr = np.array(f) 51 if arr.ndim == 3 and self.config['grayscale']: 52 arr = arr.mean(axis=2) 53 intensities.append(float(np.mean(arr))) 54 if self.config['use_motion']: 55 if prev_gray is not None: 56 diff = np.abs(arr.astype(np.float32) - prev_gray.astype(np.float32)) 57 motions.append(float(np.mean(diff))) 58 prev_gray = arr 59 fdict: Dict[str, Any] = {'name': name, 'features': {}} 60 if intensities: 61 fdict['features']['mean_intensity'] = float(np.mean(intensities)) 62 fdict['features']['std_intensity'] = float(np.std(intensities)) 63 if motions: 64 fdict['features']['motion_mean'] = float(np.mean(motions)) 65 fdict['features']['motion_std'] = float(np.std(motions)) 66 features.append(fdict) 67 return features 68 69 def get_feature_names(self) -> List[str]: 70 return ['mean_intensity', 'std_intensity', 'motion_mean', 'motion_std']
UrFall image/video feature extractor.
Extracts per-window features from sequences of images or decoded video frames:
- mean_intensity
- std_intensity
- motion_mean (if pairwise differences are available)
- motion_std
UrFallMediaFeatureExtractor(verbose: bool = False)
26 def __init__(self, verbose: bool = False): 27 super().__init__( 28 name="urfall_media", 29 description="Lightweight feature extractor for UrFall depth/RGB/video modalities" 30 ) 31 self.config = { 32 'verbose': verbose, 33 'use_motion': True, 34 'grayscale': True, # for RGB convert to gray before stats 35 }
Initialize the feature extractor.
Args: name: Name of the feature extractor description: Description of the feature extractor
def
extract_features(self, windows: List[Dict], fs: int, **kwargs) -> List[Dict]:
37 def extract_features(self, windows: List[Dict], fs: int, **kwargs) -> List[Dict]: 38 self.config.update(kwargs) 39 features: List[Dict[str, Any]] = [] 40 for window in windows: 41 name = window.get('name', 'unknown') 42 frames = window.get('data', []) 43 if not isinstance(frames, list) or len(frames) == 0: 44 continue 45 # frames is a list of numpy arrays (HxW) or (HxWxC) 46 intensities = [] 47 motions = [] 48 prev_gray = None 49 for f in frames: 50 arr = np.array(f) 51 if arr.ndim == 3 and self.config['grayscale']: 52 arr = arr.mean(axis=2) 53 intensities.append(float(np.mean(arr))) 54 if self.config['use_motion']: 55 if prev_gray is not None: 56 diff = np.abs(arr.astype(np.float32) - prev_gray.astype(np.float32)) 57 motions.append(float(np.mean(diff))) 58 prev_gray = arr 59 fdict: Dict[str, Any] = {'name': name, 'features': {}} 60 if intensities: 61 fdict['features']['mean_intensity'] = float(np.mean(intensities)) 62 fdict['features']['std_intensity'] = float(np.std(intensities)) 63 if motions: 64 fdict['features']['motion_mean'] = float(np.mean(motions)) 65 fdict['features']['motion_std'] = float(np.std(motions)) 66 features.append(fdict) 67 return features
Extract features from sliding windows.
Args: windows: List of sliding window dictionaries fs: Sampling frequency **kwargs: Additional arguments for feature extraction
Returns: List of feature dictionaries
def
get_feature_names(self) -> List[str]:
69 def get_feature_names(self) -> List[str]: 70 return ['mean_intensity', 'std_intensity', 'motion_mean', 'motion_std']
Get names of features extracted by this extractor.
Returns: List of feature names