PreProcessing.py

import numpy as np
import pandas as pd
import json
import os

#Function to remove scores from keypoints array
def remove_kpscores(keypoints):
	[keypoints.remove(s) for s in keypoints[2::3]]
	return keypoints

#Process pased obj to ake ti compatible for making df
def obj_format(obj):
	obj=obj.to_dict('index')
	key=list(obj.keys())[0]
	return obj[key]

#Function to return list of elements if path is a dir
def ret_dir(path):
	if os.path.isdir(path):
		return os.listdir(path)
	else:
		return None
		
#Main function pre-preocess pose
def preprocess_pose(pose_file):
	#Perform Pre-processing of poses
	#1. Remove extra info columns from apose results.
	#2. Keep one pose per frame with highest score.
	#Returns a dataframe with preprcessed poses
	print("PATH: ",pose_file)
	with open(pose_file) as f:
		action_sample=json.load(f)
		action_sample_new=[]
		for frame in action_sample:
			frame_obj={'image_id':frame['image_id'],
			'bbox':frame['box'],
			'score':frame['score'],
			'keypoints':remove_kpscores(frame['keypoints'])}
			action_sample_new.append(frame_obj)

		action_df=pd.DataFrame(action_sample_new) #single df with desired cols and keypoints w/o scores
		#print("Action DF: ",action_df['keypoints'][0])

		#Remove Multiple bboxes from one frame
		temp_df=action_df[['image_id','score']].groupby('image_id').max()
		
		unique_action_list=[]
		for image_id,sample in temp_df.iterrows():
			obj=(action_df[(action_df['image_id']==image_id)&(action_df['score']==sample['score'])])
			unique_action_list.append(obj_format(obj))

		unique_action_df=pd.DataFrame(unique_action_list) #df with one pose per frame, with highest score
		#print("Final df:",unique_action_df['keypoints'][0])
		print("Processed: ",pose_file)
		#This data object contains alposes, one per frame for detected persons, in following format
		# image_id - keypoints (17 xy joints) - bbox - score
		return unique_action_df

#Preprocess UPFALL poses
def upfall_poses():
	Subjects=['Subject1','Subject2','Subject3','Subject4','Subject5','Subject6','Subject7','Subject8','Subject9','Subject10','Subject11','Subject12','Subject13','Subject14','Subject15','Subject16','Subject17']
	Activities=['Activity1','Activity2','Activity3','Activity4','Activity5','Activity6','Activity7','Activity8','Activity9','Activity10','Activity11']
	Trials=['Trial1','Trial2','Trial3']
	Cameras=['Camera1','Camera2']

	#Iterate over all actions for one subject
	parent_dir='/home/mo926312/Documents/falldet/'
	poses_path=parent_dir+'poses/'
	output_dir=parent_dir+'PreProcess_poses/'

	subjects=os.listdir(poses_path)
	temp_path=''
	for sub in ['Subject6']:
		activities=ret_dir(poses_path+sub+'/')
		temp_path=poses_path+sub+'/'
		if activities is not None:
			for act in activities:
				trials=ret_dir(poses_path+sub+'/'+act+'/')
				temp_path=poses_path+sub+'/'+act+'/'
				if trials is not None:
					for trial in trials:
						cams=ret_dir(poses_path+sub+'/'+act+'/'+trial+'/')
						temp_path=poses_path+sub+'/'+act+'/'+trial+'/'
						if cams is not None:
							for cam in cams:
								cam_path=poses_path+sub+'/'+act+'/'+trial+'/'+cam+'/'
								temp_path=poses_path+sub+'/'+act+'/'+trial+'/'+cam+'/'
								if os.path.isdir(cam_path):
									activity_poses=cam_path+'alphapose-results.json'
									#Perform Pre-processing of poses
									unique_action_df = preprocess_pose(activity_poses)
									print("PATH: ",activity_poses)
									if not os.path.exists(output_dir+sub+'/'):
										os.makedirs(output_dir+sub+'/')	
									file_path=output_dir+sub+'/'+sub+act+trial+cam+'.csv'
									unique_action_df.to_csv(file_path,index=False)
			
	print("Temp Path: ",temp_path)

if __name__=="__main__":
	upfall_poses()