Dataset - CQU_TMR
Dataset - CQU_TMR
Embedded Files
This dataset consists of training and test images for Australian Road Assessment Program (AusRAP) attributes. AusRAP attributes are shown below in Figure 1.
Figure 1: AusRAP Attributes
Data Collection
Data Collection
The dataset has been prepared by extracting roadside videos provided by the Department of Transport and Main Roads (DTMR), Queensland, Australia and the Australian Road Research Board (ARRB). Thousands of real-world videos collected from a recent Digital Video Recording (DVR) survey, conducted on all state roads in Queensland, Australia were provided by industry partners, the DTMR, Queensland, and the ARRB. The images in this benchmark cover large variations in illuminance and weather conditions and include regions from different cities in Queensland, including both downtown regions and suburbs for each city. The video data contains four videos (left, right, front, rear) for each road. But only the images captured from the front camera have been used in this research. A MATLAB program is used to extract frames from video data. Each video has 500 frames, with each frame covering a 10-metre distance.
Dataset Preparation
Dataset Preparation
Samples for training, and testing were carefully selected from extracted frames avoiding consecutive scenes in these video sequences leading to many very similar images. Due to this reason, the appearances of an instance of a road safety object in the benchmark vary significantly. The road safety objects in successive shots are not related by a homography. Partially occluded or blurred objects are also included in the benchmark. To create the benchmark images, none of the images were cropped or subjected to any image processing.
Sample original images and their corresponding annotations are shown in Figure 2 below.
Figure 2: Sample Original Images and Annotated Images
Dataset Statistics
Dataset Statistics
The dataset consists of 1141 color images (image size: 1600 Ă— 1200 pixels or 1600 Ă— 1184 pixels). There are 64 classes, and 27 images per class. There are 799 training images and 342 test images. Approximately, there are 20 samples for training and 7 samples for testing each attribute. Rare attributes such as sign 10, sign 20 and sign 30 have few samples for training, while attributes such as road and lane line were present in almost all the samples. Each attribute in the dataset is manually labelled pixel-wise using Adobe Photoshop CC 2019 software. Each attribute is allocated a separate RGB color not found in the background of these images. Annotated images are .bmp images with 32 bits while original images are .jpg images.
Here are the 64 classes in the dataset:
Pole[R=255, G=0, B=0]
Tree[R=255, G=255, B=0]
Guidepost[R=0, G=255, B=150]
Flexi Post[R=255, G=160, B=100]
Supervised Crossing[R=190, G=230, B=45]
Rigid Structure[R=185, G=235, B=240]
Concrete Barrier[R=150, G=210, B=80]
Motor Cycle Friendly Barrier[R=90, G=205, B=100]
Metal Barrier[R=150, G=0, B=255]
Wire Rope Barrier[R=90, G=200, B=225]
Segregated Bicycle Path with Barrier[R=105, G=135, B=190]
Unprotected Barrier End[R=240, G=230, B=165]
Sign 10[R=255, G=60, B=90]
Sign 20[R=240, G=110, B=150]
Sign 30[R=225, G=15, B=15]
Sign 40[R=250, G=145, B=190]
Sign 50[R=240, G=200, B=240]
Sign 60[R=0, G=150, B=150]
Sign 70[R=45, G=190, B=200]
Sign 80[R=215, G=240, B=50]
Sign 90[R=250, G=130, B=90]
Sign 100[R=255, G=0, B=255]
Sign 110[R=0, G=255, B=0]
Intersection Legs - 3[R=205, G=155, B=255]
Intersection Legs - 4[R=155, G=205, B=255]
Railway Stop Signage[R=90, G=190, B=255]
Railway Sign[R=150, G=170, B=255]
Curvature Sign[R=100, G=150, B=100]
Merge Lane[R=220, G=130, B=120]
CAM[R=250, G=90, B=80]
One Way Sign[R=230, G=235, B=10]
School Zone Warning Sign[R=150, G=150, B=0]
Bicycle[R=90, G=35, B=220]
Motorcycle[R=250, G=250, B=155]
Pedestrian[R=80, G=255, B=70]
Slip Lane[R=230, G=205, B=100]
Refuge[R=45, G=165, B=65]
Median Concrete[R=50, G=120, B=180]
Median Grass[R=200, G=225, B=180]
Sidewalk - Formal[R=205, G=255, B=100]
Aggressive Vertical Face[R=225, G=165, B=125]
Large Boulder[R=15, G=130, B=165]
Upwards Slope[R=255, G=100, B=0]
Deep Drainage Ditch[R=135, G=15, B=125]
Pedestrian Fencing[R=215, G=0, B=150]
Lane Line[R=0, G=0, B=255]
Rumble Strip[R=255, G=150, B=255]
Central Hatching[R=205, G=255, B=50]
Continuous Central Turning Lane[R=100, G=255, B=150]
Left Turn Lane[R=190, G=95, B=220]
Right Turn Lane[R=100, G=60, B=10]
Parking Slot[R=190, G=245, B=245]
Pedestrian Crossing[R=0, G=150, B=0]
Bicycle Path[R=180, G=50, B=0]
Road[R=255, G=150, B=0]
Defect[R=255, G=0, B=150]
Speed Hump Sign[R=255, G=0, B=100]
Roundabout Sign[R=0, G=255, B=255]
Signed Shared Roadway[R=90, G=190, B=155]
Roadworks Sign 1[R=150, G=0, B=100]
Roadworks Sign 2[R=240, G=170, B=230]
Pedestrian Crossing Sign[R=0, G=0, B=150]
Signal Light[R=210, G=190, B=80]
Unlabeled[R=0, G=0, B=0]
Page updated
Google Sites
Report abuse