Mercurial Hosting > traffic-intelligence
view scripts/classify-objects.py @ 946:e5970606066f
bug fix on list filtering (cannot remove while iterating) and motion prediction keeping the same features
author | Nicolas Saunier <nicolas.saunier@polymtl.ca> |
---|---|
date | Fri, 21 Jul 2017 11:25:20 -0400 |
parents | 0e63a918a1ca |
children | 053484e08947 |
line wrap: on
line source
#! /usr/bin/env python import cvutils, moving, ml, storage import numpy as np import sys, argparse #from cv2.ml import SVM_RBF, SVM_C_SVC import cv2 from scipy.stats import norm, lognorm # TODO add mode detection live, add choice of kernel and svm type (to be saved in future classifier format) parser = argparse.ArgumentParser(description='The program processes indicators for all pairs of road users in the scene') parser.add_argument('--cfg', dest = 'configFilename', help = 'name of the configuration file', required = True) parser.add_argument('-d', dest = 'databaseFilename', help = 'name of the Sqlite database file (overrides the configuration file)') parser.add_argument('-i', dest = 'videoFilename', help = 'name of the video file (overrides the configuration file)') parser.add_argument('-n', dest = 'nObjects', help = 'number of objects to classify', type = int, default = None) parser.add_argument('--start-frame0', dest = 'startFrame0', help = 'starts with first frame for videos with index problem where frames cannot be reached', action = 'store_true') parser.add_argument('--plot-speed-distributions', dest = 'plotSpeedDistribution', help = 'simply plots the distributions used for each user type', action = 'store_true') parser.add_argument('--max-speed-distribution-plot', dest = 'maxSpeedDistributionPlot', help = 'if plotting the user distributions, the maximum speed to display (km/h)', type = float, default = 50.) args = parser.parse_args() params, videoFilename, databaseFilename, invHomography, intrinsicCameraMatrix, distortionCoefficients, undistortedImageMultiplication, undistort, firstFrameNum = storage.processVideoArguments(args) classifierParams = storage.ClassifierParameters(params.classifierFilename) classifierParams.convertToFrames(params.videoFrameRate, 3.6) # conversion from km/h to m/frame if classifierParams.speedAggregationMethod == 'median': speedAggregationFunc = np.median elif classifierParams.speedAggregationMethod == 'mean': speedAggregationFunc = np.mean elif classifierParams.speedAggregationMethod == 'quantile': speedAggregationFunc = lambda speeds: np.percentile(speeds, args.speedAggregationQuantile) else: print('Unknown speed aggregation method: {}. Exiting'.format(classifierParams.speedAggregationMethod)) sys.exit() pedBikeCarSVM = ml.SVM() pedBikeCarSVM.load(classifierParams.pedBikeCarSVMFilename) bikeCarSVM = ml.SVM() bikeCarSVM.load(classifierParams.bikeCarSVMFilename) # log logistic for ped and bik otherwise ((pedBeta/pedAlfa)*((sMean/pedAlfa)**(pedBeta-1)))/((1+(sMean/pedAlfa)**pedBeta)**2.) carNorm = norm(classifierParams.meanVehicleSpeed, classifierParams.stdVehicleSpeed) pedNorm = norm(classifierParams.meanPedestrianSpeed, classifierParams.stdPedestrianSpeed) # numpy lognorm shape, loc, scale: shape for numpy is scale (std of the normal) and scale for numpy is exp(location) (loc=mean of the normal) bicLogNorm = lognorm(classifierParams.scaleCyclistSpeed, loc = 0., scale = np.exp(classifierParams.locationCyclistSpeed)) speedProbabilities = {'car': lambda s: carNorm.pdf(s), 'pedestrian': lambda s: pedNorm.pdf(s), 'bicycle': lambda s: bicLogNorm.pdf(s)} if args.plotSpeedDistribution: import matplotlib.pyplot as plt plt.figure() for k in speedProbabilities: plt.plot(np.arange(0.1, args.maxSpeedDistributionPlot, 0.1), [speedProbabilities[k](s/(3.6*params.videoFrameRate)) for s in np.arange(0.1, args.maxSpeedDistributionPlot, 0.1)], label = k) maxProb = -1. for k in speedProbabilities: maxProb = max(maxProb, np.max([speedProbabilities[k](s/(3.6*params.videoFrameRate)) for s in np.arange(0.1, args.maxSpeedDistributionPlot, 0.1)])) plt.plot([classifierParams.minSpeedEquiprobable*3.6*params.videoFrameRate]*2, [0., maxProb], 'k-') plt.text(classifierParams.minSpeedEquiprobable*3.6*params.videoFrameRate, maxProb, 'threshold for equiprobable class') plt.xlabel('Speed (km/h)') plt.ylabel('Probability') plt.legend() plt.title('Probability Density Function') plt.show() sys.exit() objects = storage.loadTrajectoriesFromSqlite(databaseFilename, 'object', args.nObjects, withFeatures = True) timeInterval = moving.TimeInterval.unionIntervals([obj.getTimeInterval() for obj in objects]) if args.startFrame0: timeInterval.first = 0 capture = cv2.VideoCapture(videoFilename) width = int(capture.get(cv2.cv.CV_CAP_PROP_FRAME_WIDTH)) height = int(capture.get(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT)) #if undistort: # setup undistortion # [map1, map2] = cvutils.computeUndistortMaps(width, height, undistortedImageMultiplication, intrinsicCameraMatrix, distortionCoefficients) # height, width = map1.shape # newImgSize = (int(round(width*undistortedImageMultiplication)), int(round(height*undistortedImageMultiplication))) # newCameraMatrix = cv2.getDefaultNewCameraMatrix(intrinsicCameraMatrix, newImgSize, True) #else: # newCameraMatrix = None pastObjects = [] currentObjects = [] if capture.isOpened(): ret = True frameNum = timeInterval.first if not args.startFrame0: capture.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, frameNum) lastFrameNum = timeInterval.last while ret and frameNum <= lastFrameNum: ret, img = capture.read() if ret: if frameNum%50 == 0: print('frame number: {}'.format(frameNum)) #if undistort: # img = cv2.remap(img, map1, map2, interpolation=cv2.INTER_LINEAR) for obj in objects: if obj.getFirstInstant() <= frameNum: # if images are skipped obj.initClassifyUserTypeHoGSVM(speedAggregationFunc, pedBikeCarSVM, bikeCarSVM, classifierParams.maxPedestrianSpeed, classifierParams.maxCyclistSpeed, classifierParams.nFramesIgnoreAtEnds, invHomography, intrinsicCameraMatrix, distortionCoefficients) currentObjects.append(obj) objects[:] = [obj for obj in objects if obj.getFirstInstant() > frameNum] for obj in currentObjects: if obj.getLastInstant() <= frameNum: # if images are skipped obj.classifyUserTypeHoGSVM(minSpeedEquiprobable = classifierParams.minSpeedEquiprobable, speedProbabilities = speedProbabilities, maxPercentUnknown = classifierParams.maxPercentUnknown) pastObjects.append(obj) else: obj.classifyUserTypeHoGSVMAtInstant(img, frameNum, width, height, classifierParams.percentIncreaseCrop, classifierParams.percentIncreaseCrop, classifierParams.minNPixels, classifierParams.hogRescaleSize, classifierParams.hogNOrientations, classifierParams.hogNPixelsPerCell, classifierParams.hogNCellsPerBlock, classifierParams.hogBlockNorm) currentObjects[:] = [obj for obj in objects if obj.getLastInstant() > frameNum] frameNum += 1 for obj in currentObjects: obj.classifyUserTypeHoGSVM(minSpeedEquiprobable = classifierParams.minSpeedEquiprobable, speedProbabilities = speedProbabilities, maxPercentUnknown = classifierParams.maxPercentUnknown) pastObjects.append(obj) print('Saving user types') storage.setRoadUserTypes(databaseFilename, pastObjects)