Mercurial Hosting > traffic-intelligence
view scripts/dltrack.py @ 1237:31a441efca6c
ped-bike grouping working, bike merging left todo
author | Nicolas Saunier <nicolas.saunier@polymtl.ca> |
---|---|
date | Mon, 02 Oct 2023 16:51:43 -0400 |
parents | 100fe098abe9 |
children | b684135d817f |
line wrap: on
line source
#! /usr/bin/env python3 # from https://docs.ultralytics.com/modes/track/ import sys, argparse from copy import copy from collections import Counter import numpy as np from scipy.optimize import linear_sum_assignment from ultralytics import YOLO from torch import cat from torchvision.ops import box_iou import cv2 from trafficintelligence import cvutils, moving, storage, utils parser = argparse.ArgumentParser(description='The program tracks objects following the ultralytics yolo executable.')#, epilog = 'Either the configuration filename or the other parameters (at least video and database filenames) need to be provided.') parser.add_argument('-i', dest = 'videoFilename', help = 'name of the video file', required = True) parser.add_argument('-d', dest = 'databaseFilename', help = 'name of the Sqlite database file', required = True) parser.add_argument('-m', dest = 'detectorFilename', help = 'name of the detection model file', required = True) parser.add_argument('-t', dest = 'trackerFilename', help = 'name of the tracker file', required = True) parser.add_argument('--display', dest = 'display', help = 'show the results (careful with long videos, risk of running out of memory)', action = 'store_true') parser.add_argument('-f', dest = 'firstFrameNum', help = 'number of first frame number to process', type = int, default = 0) parser.add_argument('-l', dest = 'lastFrameNum', help = 'number of last frame number to process', type = int, default = float('Inf')) parser.add_argument('--bike-prop', dest = 'bikeProportion', help = 'minimum proportion of time a person classified as bike or motorbike to be classified as cyclist', type = float, default = 0.2) parser.add_argument('--cyclist-iou', dest = 'cyclistIou', help = 'IoU threshold to associate a bike and ped bounding box', type = float, default = 0.15) parser.add_argument('--cyclist-match-prop', dest = 'cyclistMatchingProportion', help = 'minimum proportion of time a bike exists and is associated with a pedestrian to be merged as cyclist', type = float, default = 0.3) # mask!! args = parser.parse_args() # required functionality? # # filename of the video to process (can be images, eg image%04d.png) # video-filename = laurier.avi # # filename of the database where results are saved # database-filename = laurier.sqlite # # filename of the homography matrix # homography-filename = laurier-homography.txt # # filename of the camera intrinsic matrix # intrinsic-camera-filename = intrinsic-camera.txt # # -0.11759321 0.0148536 0.00030756 -0.00020578 -0.00091816 # distortion-coefficients = -0.11759321 # distortion-coefficients = 0.0148536 # distortion-coefficients = 0.00030756 # distortion-coefficients = -0.00020578 # distortion-coefficients = -0.00091816 # # undistorted image multiplication # undistorted-size-multiplication = 1.31 # # Interpolation method for remapping image when correcting for distortion: 0 for INTER_NEAREST - a nearest-neighbor interpolation; 1 for INTER_LINEAR - a bilinear interpolation (used by default); 2 for INTER_CUBIC - a bicubic interpolation over 4x4 pixel neighborhood; 3 for INTER_LANCZOS4 # interpolation-method = 1 # # filename of the mask image (where features are detected) # mask-filename = none # # undistort the video for feature tracking # undistort = false # # load features from database # load-features = false # # display trajectories on the video # display = false # # original video frame rate (number of frames/s) # video-fps = 29.97 # # number of digits of precision for all measurements derived from video # # measurement-precision = 3 # # first frame to process # frame1 = 0 # # number of frame to process: 0 means processing all frames # nframes = 0 # TODO add option to refine position with mask for vehicles # use 2 x bytetrack track buffer to remove objects from existing ones # check if one can go to specific frame https://docs.ultralytics.com/modes/track/#persisting-tracks-loop # Load a model model = YOLO(args.detectorFilename) # seg yolov8x-seg.pt # seg could be used on cropped image... if can be loaded and kept in memory # model = YOLO('/home/nicolas/Research/Data/classification-models/yolo_nas_l.pt ') # AttributeError: 'YoloNAS_L' object has no attribute 'get' # Track with the model if args.display: windowName = 'frame' cv2.namedWindow(windowName, cv2.WINDOW_NORMAL) capture = cv2.VideoCapture(args.videoFilename) #results = model.track(source=args.videoFilename, tracker="/home/nicolas/Research/Data/classification-models/bytetrack.yaml", classes=list(moving.cocoTypeNames.keys()), stream=True) objects = [] currentObjects = {} featureNum = 0 frameNum = args.firstFrameNum capture.set(cv2.CAP_PROP_POS_FRAMES, frameNum) lastFrameNum = args.lastFrameNum success, frame = capture.read() results = model.track(frame, tracker=args.trackerFilename, classes=list(moving.cocoTypeNames.keys()), persist=True, verbose=False) # create object with user type and list of 3 features (bottom ones and middle) + projection while capture.isOpened() and success and frameNum <= lastFrameNum: #for frameNum, result in enumerate(results): result = results[0] print(frameNum, len(result.boxes), 'objects') for box in result.boxes: #print(box.cls, box.id, box.xyxy) if box.id is not None: # None are objects with low confidence num = int(box.id.item()) #xyxy = box.xyxy[0].tolist() if num in currentObjects: currentObjects[num].timeInterval.last = frameNum currentObjects[num].bboxes[frameNum] = copy(box.xyxy) currentObjects[num].userTypes.append(moving.coco2Types[int(box.cls.item())]) currentObjects[num].features[0].tmpPositions[frameNum] = moving.Point(box.xyxy[0,0].item(), box.xyxy[0,1].item()) currentObjects[num].features[1].tmpPositions[frameNum] = moving.Point(box.xyxy[0,2].item(), box.xyxy[0,3].item()) else: inter = moving.TimeInterval(frameNum,frameNum) currentObjects[num] = moving.MovingObject(num, inter) currentObjects[num].bboxes = {frameNum: copy(box.xyxy)} currentObjects[num].userTypes = [moving.coco2Types[int(box.cls.item())]] currentObjects[num].features = [moving.MovingObject(featureNum), moving.MovingObject(featureNum+1)] currentObjects[num].featureNumbers = [featureNum, featureNum+1] currentObjects[num].features[0].tmpPositions = {frameNum: moving.Point(box.xyxy[0,0].item(), box.xyxy[0,1].item())} currentObjects[num].features[1].tmpPositions = {frameNum: moving.Point(box.xyxy[0,2].item(), box.xyxy[0,3].item())} featureNum += 2 if args.display: cvutils.cvImshow(windowName, result.plot()) # original image in orig_img key = cv2.waitKey() if cvutils.quitKey(key): break frameNum += 1 success, frame = capture.read() results = model.track(frame, persist=True) # classification for num, obj in currentObjects.items(): #obj.setUserType(utils.mostCommon(obj.userTypes)) # improve? mix with speed? userTypeStats = Counter(obj.userTypes) if (4 in userTypeStats or (3 in userTypeStats and 4 in userTypeStats and userTypeStats[3]<=userTypeStats[4])) and userTypeStats[3]+userTypeStats[4] > args.bikeProportion*userTypeStats.total(): # 3 is motorcycle and 4 is cyclist (verif if not turning all motorbike into cyclists) obj.setUserType(4) else: obj.setUserType(userTypeStats.most_common()[0][0]) # merge bikes and people twowheels = [num for num, obj in currentObjects.items() if obj.getUserType() in (3,4)] pedestrians = [num for num, obj in currentObjects.items() if obj.getUserType() == 2] costs = [] for twInd in twowheels: tw = currentObjects[twInd] twCost = [] for pedInd in pedestrians: ped = currentObjects[pedInd] nmatches = 0 for t in tw.bboxes: if t in ped.bboxes: #print(tw.num, ped.num, t, box_iou(tw.bboxes[t], ped.bboxes[t])) if box_iou(tw.bboxes[t], ped.bboxes[t]).item() > args.cyclistIou: nmatches += 1 twCost.append(nmatches/len(tw.bboxes)) costs.append(twCost) costs = -np.array(costs) # before matching, scan for pedestrians with good non-overlapping temporal match with different bikes for pedInd in costs.shape[1]: if sum(costs[:,pedInd] < -args.cyclistMatchingProportion) >1: twIndices = np.nonzero(costs[:,pedInd] < -args.cyclistMatchingProportion) # we have to compute temporal overlaps with everyone else, then remove the ones with the most overlap (sum over column) one by one until there is little left temporalOverlaps = np.zeros((len(twIndices),len(twIndices))) twIndices, matchingPedIndices = linear_sum_assignment(costs) for twInd, pedInd in zip(twIndices, matchingPedIndices): # caution indices in the cost matrix if -costs[twInd, pedInd] >= args.cyclistMatchingProportion: tw = currentObjects[twowheels[twInd]] ped = currentObjects[pedestrians[pedInd]] timeInstants = set(tw.bboxes).union(set(ped.bboxes)) for t in timeInstants: if t in tw.bboxes and t in ped.bboxes: tw.features[0].tmpPositions[t] = moving.Point(min(tw.features[0].tmpPositions[t].x, ped.features[0].tmpPositions[t].x), min(tw.features[0].tmpPositions[t].y, ped.features[0].tmpPositions[t].y)) tw.features[1].tmpPositions[t] = moving.Point(max(tw.features[1].tmpPositions[t].x, ped.features[1].tmpPositions[t].x), max(tw.features[1].tmpPositions[t].y, ped.features[1].tmpPositions[t].y)) elif t in ped.bboxes: tw.features[0].tmpPositions[t] = ped.features[0].tmpPositions[t] tw.features[1].tmpPositions[t] = ped.features[1].tmpPositions[t] tw.timeInterval = moving.TimeInterval(min(tw.getFirstInstant(), ped.getFirstInstant()), max(tw.getLastInstant(), ped.getLastInstant())) del currentObjects[pedestrians[pedInd]] #Verif overlap piéton vélo : si long hors overlap, changement mode (trouver exemples) # interpolate and generate velocity (?) before saving for num, obj in currentObjects.items(): obj.features[0].timeInterval = copy(obj.getTimeInterval()) obj.features[1].timeInterval = copy(obj.getTimeInterval()) if obj.length() != len(obj.features[0].tmpPositions): # interpolate obj.features[0].positions = moving.Trajectory.fromPointDict(obj.features[0].tmpPositions) obj.features[1].positions = moving.Trajectory.fromPointDict(obj.features[1].tmpPositions) else: obj.features[0].positions = moving.Trajectory.fromPointList(list(obj.features[0].tmpPositions.values())) obj.features[1].positions = moving.Trajectory.fromPointList(list(obj.features[1].tmpPositions.values())) storage.saveTrajectoriesToSqlite(args.databaseFilename, list(currentObjects.values()), 'object') # todo save bbox and mask to study localization / representation # apply quality checks deviation and acceleration bounds?