Mercurial Hosting > traffic-intelligence
view scripts/process.py @ 1075:67144f26609e
Updates crop
author | Wendlasida |
---|---|
date | Wed, 11 Jul 2018 15:01:37 -0400 |
parents | c9c03c97ed9f |
children | 9d4a06f49cb8 |
line wrap: on
line source
#! /usr/bin/env python3 import sys, argparse from pathlib import Path from multiprocessing.pool import Pool #import matplotlib #atplotlib.use('Agg') import matplotlib.pyplot as plt import numpy as np from pandas import DataFrame from trafficintelligence import storage, events, prediction, cvutils, utils, moving, processing, ml from trafficintelligence.metadata import * parser = argparse.ArgumentParser(description='This program manages the processing of several files based on a description of the sites and video data in an SQLite database following the metadata module.') # input parser.add_argument('--db', dest = 'metadataFilename', help = 'name of the metadata file', required = True) parser.add_argument('--videos', dest = 'videoIds', help = 'indices of the video sequences', nargs = '*', type = int) parser.add_argument('--sites', dest = 'siteIds', help = 'indices of the video sequences', nargs = '*') # main function parser.add_argument('--delete', dest = 'delete', help = 'data to delete', choices = ['feature', 'object', 'classification', 'interaction']) parser.add_argument('--process', dest = 'process', help = 'data to process', choices = ['feature', 'object', 'classification', 'prototype', 'interaction']) parser.add_argument('--display', dest = 'display', help = 'data to display (replay over video)', choices = ['feature', 'object', 'classification', 'interaction']) parser.add_argument('--analyze', dest = 'analyze', help = 'data to analyze (results)', choices = ['feature', 'object', 'classification', 'interaction']) # common options parser.add_argument('--cfg', dest = 'configFilename', help = 'name of the configuration file') parser.add_argument('-n', dest = 'nObjects', help = 'number of objects/interactions to process', type = int) parser.add_argument('-t', dest = 'trajectoryType', help = 'type of trajectories', choices = ['feature', 'object'], default = 'feature') parser.add_argument('--dry', dest = 'dryRun', help = 'dry run of processing', action = 'store_true') parser.add_argument('--nthreads', dest = 'nProcesses', help = 'number of processes to run in parallel', type = int, default = 1) parser.add_argument('--subsample', dest = 'positionSubsamplingRate', help = 'rate of position subsampling (1 every n positions)', type = int) ### process options # motion pattern learning and assignment parser.add_argument('--prototype-filename', dest = 'outputPrototypeDatabaseFilename', help = 'name of the Sqlite database file to save prototypes', default = 'prototypes.sqlite') #parser.add_argument('-i', dest = 'inputPrototypeDatabaseFilename', help = 'name of the Sqlite database file for prototypes to start the algorithm with') parser.add_argument('--nobjects-mp', dest = 'nMPObjects', help = 'number of objects/interactions to process', type = int) parser.add_argument('--nfeatures-per-object', dest = 'nLongestFeaturesPerObject', help = 'maximum number of features per object to load', type = int) parser.add_argument('--epsilon', dest = 'epsilon', help = 'distance for the similarity of trajectory points', type = float) parser.add_argument('--metric', dest = 'metric', help = 'metric for the similarity of trajectory points', default = 'cityblock') # default is manhattan distance parser.add_argument('--minsimil', dest = 'minSimilarity', help = 'minimum similarity to put a trajectory in a cluster', type = float) parser.add_argument('--min-cluster-size', dest = 'minClusterSize', help = 'minimum cluster size', type = int, default = 0) #parser.add_argument('--learn', dest = 'learn', help = 'learn', action = 'store_true') parser.add_argument('--optimize', dest = 'optimizeCentroid', help = 'recompute centroid at each assignment', action = 'store_true') parser.add_argument('--random', dest = 'randomInitialization', help = 'random initialization of clustering algorithm', action = 'store_true') #parser.add_argument('--similarities-filename', dest = 'similaritiesFilename', help = 'filename of the similarities') parser.add_argument('--save-similarities', dest = 'saveSimilarities', help = 'save computed similarities (in addition to prototypes)', action = 'store_true') parser.add_argument('--save-assignments', dest = 'saveAssignments', help = 'saves the assignments of the objects to the prototypes', action = 'store_true') parser.add_argument('--assign', dest = 'assign', help = 'assigns the objects to the prototypes and saves the assignments', action = 'store_true') # safety analysis parser.add_argument('--prediction-method', dest = 'predictionMethod', help = 'prediction method (constant velocity (cvd: vector computation (approximate); cve: equation solving; cv: discrete time (approximate)), normal adaptation, point set prediction)', choices = ['cvd', 'cve', 'cv', 'na', 'ps', 'mp']) parser.add_argument('--pet', dest = 'computePET', help = 'computes PET', action = 'store_true') # override other tracking config, erase sqlite? # analysis options parser.add_argument('--output', dest = 'output', help = 'kind of output to produce (interval means)', choices = ['figure', 'interval', 'event']) parser.add_argument('--min-user-duration', dest = 'minUserDuration', help = 'mininum duration we have to see the user to take into account in the analysis (s)', type = float, default = 0.1) parser.add_argument('--interval-duration', dest = 'intervalDuration', help = 'length of time interval to aggregate data (min)', type = float, default = 15.) parser.add_argument('--aggregation', dest = 'aggMethod', help = 'aggregation method per user/event and per interval', choices = ['mean', 'median', 'centile'], nargs = '*', default = ['median']) parser.add_argument('--aggregation-centile', dest = 'aggCentiles', help = 'centile(s) to compute from the observations', nargs = '*', type = int) dpi = 150 # unit of analysis: site or video sequence? # need way of selecting sites as similar as possible to sql alchemy syntax # override tracking.cfg from db # manage cfg files, overwrite them (or a subset of parameters) # delete sqlite files # info of metadata args = parser.parse_args() ################################# # Data preparation ################################# session = connectDatabase(args.metadataFilename) parentPath = Path(args.metadataFilename).parent # files are relative to metadata location videoSequences = [] sites = [] if args.videoIds is not None: videoSequences = [session.query(VideoSequence).get(videoId) for videoId in args.videoIds] siteIds = set([vs.cameraView.siteIdx for vs in videoSequences]) elif args.siteIds is not None: siteIds = set(args.siteIds) for siteId in siteIds: tmpsites = getSite(session, siteId) sites.extend(tmpsites) for site in tmpsites: for cv in site.cameraViews: videoSequences.extend(cv.videoSequences) else: print('No video/site to process') if args.nProcesses > 1: pool = Pool(args.nProcesses) ################################# # Delete ################################# if args.delete is not None: if args.delete == 'feature': response = input('Are you sure you want to delete the tracking results (SQLite files) of all these sites (y/n)?') if response == 'y': for vs in videoSequences: p = parentPath.absolute()/vs.getDatabaseFilename() p.unlink() elif args.delete in ['object', 'interaction']: #parser.add_argument('-t', dest = 'dataType', help = 'type of the data to remove', required = True, choices = ['object','interaction', 'bb', 'pois', 'prototype']) for vs in videoSequences: storage.deleteFromSqlite(str(parentPath/vs.getDatabaseFilename()), args.delete) ################################# # Process ################################# if args.process in ['feature', 'object']: # tracking if args.nProcesses == 1: for vs in videoSequences: if not (parentPath/vs.getDatabaseFilename()).exists() or args.process == 'object': if args.configFilename is None: configFilename = str(parentPath/vs.cameraView.getTrackingConfigurationFilename()) else: configFilename = args.configFilename if vs.cameraView.cameraType is None: cvutils.tracking(configFilename, args.process == 'object', str(parentPath.absolute()/vs.getVideoSequenceFilename()), str(parentPath.absolute()/vs.getDatabaseFilename()), str(parentPath.absolute()/vs.cameraView.getHomographyFilename()), str(parentPath.absolute()/vs.cameraView.getMaskFilename()), False, None, None, args.dryRun) else: cvutils.tracking(configFilename, args.process == 'object', str(parentPath.absolute()/vs.getVideoSequenceFilename()), str(parentPath.absolute()/vs.getDatabaseFilename()), str(parentPath.absolute()/vs.cameraView.getHomographyFilename()), str(parentPath.absolute()/vs.cameraView.getMaskFilename()), True, vs.cameraView.cameraType.intrinsicCameraMatrix, vs.cameraView.cameraType.distortionCoefficients, args.dryRun) else: print('SQLite already exists: {}'.format(parentPath/vs.getDatabaseFilename())) else: for vs in videoSequences: if not (parentPath/vs.getDatabaseFilename()).exists() or args.process == 'object': if args.configFilename is None: configFilename = str(parentPath/vs.cameraView.getTrackingConfigurationFilename()) else: configFilename = args.configFilename if vs.cameraView.cameraType is None: pool.apply_async(cvutils.tracking, args = (configFilename, args.process == 'object', str(parentPath.absolute()/vs.getVideoSequenceFilename()), str(parentPath.absolute()/vs.getDatabaseFilename()), str(parentPath.absolute()/vs.cameraView.getHomographyFilename()), str(parentPath.absolute()/vs.cameraView.getMaskFilename()), False, None, None, args.dryRun)) else: pool.apply_async(cvutils.tracking, args = (configFilename, args.process == 'object', str(parentPath.absolute()/vs.getVideoSequenceFilename()), str(parentPath.absolute()/vs.getDatabaseFilename()), str(parentPath.absolute()/vs.cameraView.getHomographyFilename()), str(parentPath.absolute()/vs.cameraView.getMaskFilename()), True, vs.cameraView.cameraType.intrinsicCameraMatrix, vs.cameraView.cameraType.distortionCoefficients, args.dryRun)) else: print('SQLite already exists: {}'.format(parentPath/vs.getDatabaseFilename())) pool.close() pool.join() elif args.process == 'prototype': # motion pattern learning # learn by site by default -> group videos by site (or by camera view? TODO add cameraviews) # by default, load all objects, learn and then assign (BUT not save the assignments) for site in sites: print('Learning motion patterns for site {} ({})'.format(site.idx, site.name)) objects = {} object2VideoSequences = {} for cv in site.cameraViews: for vs in cv.videoSequences: print('Loading '+vs.getDatabaseFilename()) objects[vs.idx] = storage.loadTrajectoriesFromSqlite(str(parentPath/vs.getDatabaseFilename()), args.trajectoryType, args.nObjects, timeStep = args.positionSubsamplingRate, nLongestFeaturesPerObject = args.nLongestFeaturesPerObject) if args.trajectoryType == 'object' and args.nLongestFeaturesPerObject is not None: objectsWithFeatures = objects[vs.idx] objects[vs.idx] = [f for o in objectsWithFeatures for f in o.getFeatures()] prototypeType = 'feature' else: prototypeType = args.trajectoryType for obj in objects[vs.idx]: object2VideoSequences[obj] = vs lcss = utils.LCSS(metric = args.metric, epsilon = args.epsilon) similarityFunc = lambda x,y : lcss.computeNormalized(x, y) trainingObjects = [o for tmpobjects in objects.values() for o in tmpobjects] if args.nMPObjects is not None and args.nMPObjects < len(trainingObjects): m = int(np.floor(float(len(trainingObjects))/args.nMPObjects)) trainingObjects = trainingObjects[::m] similarities = -np.ones((len(trainingObjects), len(trainingObjects))) prototypeIndices, labels = processing.learnAssignMotionPatterns(True, True, trainingObjects, similarities, args.minSimilarity, similarityFunc, args.minClusterSize, args.optimizeCentroid, args.randomInitialization, True, []) if args.outputPrototypeDatabaseFilename is None: outputPrototypeDatabaseFilename = args.databaseFilename else: outputPrototypeDatabaseFilename = args.outputPrototypeDatabaseFilename # TODO maintain mapping from object prototype to db filename + compute nmatchings before clusterSizes = ml.computeClusterSizes(labels, prototypeIndices, -1) storage.savePrototypesToSqlite(str(parentPath/site.getPath()/outputPrototypeDatabaseFilename), [moving.Prototype(object2VideoSequences[trainingObjects[i]].getDatabaseFilename(False), trainingObjects[i].getNum(), prototypeType, clusterSizes[i]) for i in prototypeIndices]) elif args.process == 'interaction': # safety analysis TODO make function in safety analysis script if args.predictionMethod == 'cvd': predictionParameters = prediction.CVDirectPredictionParameters() if args.predictionMethod == 'cve': predictionParameters = prediction.CVExactPredictionParameters() for vs in videoSequences: print('Processing '+vs.getDatabaseFilename()) objects = storage.loadTrajectoriesFromSqlite(str(parentPath/vs.getDatabaseFilename()), 'object')#, args.nObjects, withFeatures = (params.useFeaturesForPrediction or predictionMethod == 'ps' or predictionMethod == 'mp')) interactions = events.createInteractions(objects) #if args.nProcesses == 1: #print(str(parentPath/vs.cameraView.getTrackingConfigurationFilename())) params = storage.ProcessParameters(str(parentPath/vs.cameraView.getTrackingConfigurationFilename())) #print(len(interactions), args.computePET, predictionParameters, params.collisionDistance, params.predictionTimeHorizon, params.crossingZones) processed = events.computeIndicators(interactions, True, args.computePET, predictionParameters, params.collisionDistance, params.predictionTimeHorizon, params.crossingZones, False, None) storage.saveIndicatorsToSqlite(str(parentPath/vs.getDatabaseFilename()), processed) # else: # pool = Pool(processes = args.nProcesses) # nInteractionPerProcess = int(np.ceil(len(interactions)/float(args.nProcesses))) # jobs = [pool.apply_async(events.computeIndicators, args = (interactions[i*nInteractionPerProcess:(i+1)*nInteractionPerProcess], not args.noMotionPrediction, args.computePET, predictionParameters, params.collisionDistance, params.predictionTimeHorizon, params.crossingZones, False, None)) for i in range(args.nProcesses)] # processed = [] # for job in jobs: # processed += job.get() # pool.close() ################################# # Analyze ################################# if args.analyze == 'object': # user speeds, accelerations # aggregation per site data = [] # list of observation per site-user with time headers = ['sites', 'date', 'time', 'user_type'] aggFunctions = {} for method in args.aggMethod: if method == 'centile': aggFunctions[method] = utils.aggregationFunction(method, args.aggCentiles) for c in args.aggCentiles: headers.append('{}{}'.format(method,c)) else: aggFunctions[method] = utils.aggregationFunction(method) headers.append(method) for vs in videoSequences: d = vs.startTime.date() t1 = vs.startTime.time() minUserDuration = args.minUserDuration*vs.cameraView.cameraType.frameRate print('Extracting speed from '+vs.getDatabaseFilename()) objects = storage.loadTrajectoriesFromSqlite(str(parentPath/vs.getDatabaseFilename()), 'object', args.nObjects) for o in objects: if o.length() > minUserDuration: row = [vs.cameraView.siteIdx, d, utils.framesToTime(o.getFirstInstant(), vs.cameraView.cameraType.frameRate, t1), o.getUserType()] tmp = o.getSpeeds() for method,func in aggFunctions.items(): aggSpeeds = vs.cameraView.cameraType.frameRate*3.6*func(tmp) if method == 'centile': row += aggSpeeds.tolist() else: row.append(aggSpeeds) data.append(row) data = DataFrame(data, columns = headers) if args.output == 'figure': for name in headers[4:]: plt.ioff() plt.figure() plt.boxplot([data.loc[data['sites']==siteId, name] for siteId in siteIds], labels = [session.query(Site).get(siteId).name for siteId in siteIds]) plt.ylabel(name+' Speeds (km/h)') plt.savefig(name.lower()+'-speeds.png', dpi=dpi) plt.close() elif args.output == 'event': data.to_csv('speeds.csv', index = False) if args.analyze == 'interaction': indicatorIds = [2,5,7,10] conversionFactors = {2: 1., 5: 30.*3.6, 7:1./30, 10:1./30} maxIndicatorValue = {2: float('inf'), 5: float('inf'), 7:10., 10:10.} indicators = {} interactions = {} for vs in videoSequences: if not vs.cameraView.siteIdx in interactions: interactions[vs.cameraView.siteIdx] = [] indicators[vs.cameraView.siteIdx] = {} for i in indicatorIds: indicators[vs.cameraView.siteIdx][i] = [] interactions[vs.cameraView.siteIdx] += storage.loadInteractionsFromSqlite(str(parentPath/vs.getDatabaseFilename())) print(vs.getDatabaseFilename(), len(interactions[vs.cameraView.siteIdx])) for inter in interactions[vs.cameraView.siteIdx]: for i in indicatorIds: indic = inter.getIndicator(events.Interaction.indicatorNames[i]) if indic is not None: v = indic.getMostSevereValue()*conversionFactors[i] if v < maxIndicatorValue[i]: indicators[vs.cameraView.siteIdx][i].append(v) for i in indicatorIds: tmp = [indicators[siteId][i] for siteId in indicators] plt.ioff() plt.figure() plt.boxplot(tmp, labels = [session.query(Site).get(siteId).name for siteId in indicators]) plt.ylabel(events.Interaction.indicatorNames[i]+' ('+events.Interaction.indicatorUnits[i]+')') plt.savefig(events.Interaction.indicatorNames[i]+'.png', dpi=150) plt.close()