Automated lane boundary annotations for the Malaysian urban and highway driving dataset

Description

This dataset provides automated, per frame lane boundary annotations for the 27 video clips released in the companion Malaysian Urban and Highway Driving Dataset by the same author on Mendeley Data: Clip #1: https://doi.org/10.17632/rkyfwv9hw8.1 Clip #2: https://doi.org/10.17632/ynmympptmy.1 Clip #3: https://doi.org/10.17632/m88955h7vb.1 Clip #4: https://doi.org/10.17632/vry8n5nkrf.1 Clip #5: https://doi.org/10.17632/f24x2p6b5h.3 Clip #6: https://doi.org/10.17632/xskxs82mz6.3 Clip #7: https://doi.org/10.17632/dppstzh8n6.4 Clip #8: https://doi.org/10.17632/hgt5whhj6n.3 Clip #9: https://doi.org/10.17632/bvbykc4hxf.4 Clip #10: https://doi.org/10.17632/g98zzcn6nr.3 Clip #11: https://doi.org/10.17632/z3yjbd4567.3 Clip #12: https://doi.org/10.17632/ytn823rw8j.3 Clip #13: https://doi.org/10.17632/946jzttn7n.3 Clip #14: https://doi.org/10.17632/cww75348bj.3 Clip #15: https://doi.org/10.17632/k74tdgbhjm.3 Clip #16: https://doi.org/10.17632/hps9jsjwxp.4 Clip #17: https://doi.org/10.17632/bxmmttx535.3 Clip #18: https://doi.org/10.17632/smx7tbx29p.3 Clip #19: https://doi.org/10.17632/kcxpm835gw.3 Clip #20: https://doi.org/10.17632/m25z57438h.3 Clip #21: https://doi.org/10.17632/cjptbmddpk.4 Clip #22: https://doi.org/10.17632/yhd2j7ddxc.3 Clip #23: https://doi.org/10.17632/5zjf62drv7.3 Clip #24: https://doi.org/10.17632/r8vm7nbgvm.3 Clip #25: https://doi.org/10.17632/642n3xx8s6.3 Clip #26: https://doi.org/10.17632/wmymrk79tg.3 Clip #27: https://doi.org/10.17632/wb4hgnr6k3.3 Annotations were generated using YOLOP, a pretrained deep learning model for lane and drivable area segmentation, run at 5 frames per second across all 27 clips, giving 11,665 annotated frames in total. For each sampled frame, the release provides: pixel coordinates of the top and bottom points of the left and right ego lane boundary; a lateral offset ratio and derived lane departure flag, computed using the same definition as the vision based lane departure warning framework validated on this dataset in Poh Ping et al. (2019, Heliyon, https://doi.org/10.1016/j.heliyon.2019.e02169); a detection_status field recording whether both, one, or neither lane boundary was found in that frame; the same annotations in TuSimple format for direct use in common lane detection training pipelines; the corresponding extracted JPEG frame for every annotated sample; a set of quality assurance overlay images for visual auditing; and a metadata file documenting the full method, licensing, sampling scheme, and field definitions. Across the 11,665 annotated frames, both lane boundaries were detected in 99.8 percent of frames. These are silver labels: they are generated automatically and have not been manually verified, and should be treated as a starting point for training or benchmarking rather than as ground truth. Full detection quality statistics and known limitations are documented in the included metadata file. Released under CC BY 4.0, consistent with the rest of the dataset.

Steps to reproduce

1. Obtain the pretrained YOLOP lane segmentation weights (yolop-320-320.onnx) from the official repository (https://github.com/hustvl/YOLOP), used unmodified. Run inference with ONNX Runtime; only the lane_line_seg output head is used. 2. For each video frame to be annotated, convert from BGR to RGB, letterbox resize to 320 by 320 pixels preserving aspect ratio (grey padding), and normalize using ImageNet mean 0.485, 0.456, 0.406 and standard deviation 0.229, 0.224, 0.225. 3. Run the normalized frame through the model. Take the argmax of the two channel lane segmentation output to obtain a binary lane mask, crop out the letterbox padding, and resize the mask back to the frame's native resolution (320 by 180 for Clips #1 to #4, 320 by 176 for Clips #5 to #27). 4. Sample the mask at 18 fixed row heights (y = 88 to 173 in steps of 5). At each row, group lane pixels into connected blobs and assign the blob nearest the frame's horizontal centre to the left or right ego lane boundary. 5. Compute the lateral offset ratio as the smaller of the two distances, right boundary to centre or centre to left boundary, minus TH times the centre distance, divided by TH times the centre distance. Use the bottommost valid point on each boundary, Xm as the frame's horizontal centre, and TH equal to 0.8, consistent with the definition in Poh Ping et al. 2019, Heliyon, https://doi.org/10.1016/j.heliyon.2019.e02169. Set lane_departure_flag to 1 when this ratio is less than or equal to 0. 6. Repeat steps 2 to 5 for every sixth frame of each clip, 5 frames per second sampled from a native 30 frames per second recording, across all 27 clips. Write one row per sampled frame to the per clip and combined CSV files, and one JSON line per sampled frame to the combined TuSimple format file. 7. Extract and save the corresponding frame as a JPEG image for every sampled frame. For the first, middle, and last sampled frame of each clip, additionally save a quality assurance overlay image showing the raw segmentation mask and the derived left and right lane points superimposed on the frame.

Institutions

• Multimedia University

  Melaka, Malacca

Categories

Licence