Perception presentation

cu icar: perception course

1 3D recognition of lidar clound data:detect pedestrians and bicyclists

original paper:http://openaccess.thecvf.com/content_cvpr_2018/CameraReady/3333.pdf

Voxel partition(3d space divided into equally spaced voxels  D*H*W)--grouping--random sampling

their presentation is similar to ref:https://blog.csdn.net/pikachu_777/article/details/83000934

T samples points from voxel  are sent to full connected layer

KITTI dataset  :http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d   you can got the 3d lidar data here.

chanllenges: distant objects may not be detected due to sparse spoint clounds.

Related Datasets

 

• CERV Vehicle Lights Dataset: Annotations of vehicle lights for a subset of the object detection benchmark.
• PASCAL3D+: Augments 12 rigid object classes of PASCAL VOC 2012 with 3D annotations.
• The PASCAL Visual Object Classes Challenges: Dataset and benchmarks for object class recognition.
• TME Motorway Dataset: 28 video sequences with vehicle annotations captured from VisLab's BRAiVE vehicle.
• LabelMe: Online annotation tool to build image databases for computer vision research.
• MIT Street Scenes: Street-side images with labels for 9 object categories (including cars, pedestrians, buildings, trees).
• Daimler Pedestrian Datasets: Datasets focusing on pedestrian detection for autonomous driving.
• Caltech Pedestrian Detection Benchmark: 10 hours of video with 350.000 annotated pedestrian bounding boxes.
• Robust Multi-Person Tracking from Mobile Platforms: Videos with annotated pedestrians captured from a stroller

2 pedestrians detection using setero vision

motivation:understand hunman view single object

goals: model the stereo camera and RGBD camera

existing solutin:stereo matching, estimation, segmentation,/objecct detection and surface normal vector estimation

NERV object detection

issues:lighting,lacks robustness,

camera:https://www.extremetech.com/computing/280535-intels-new-realsense-d435i-adds-imu-for-integrated-tracking

3 lidar -camera calibration

assign RBG piexel to lidar point clound

use rgb information to color the object.

coordinates: different orientations, /diplaced ceter point;sensors at different locations on vehicle

use triagle as object.

more points on the triangle,large enought to camputre enought lidar data.

KIT dataset 

calibrate the camera and lidar. : according to the measured point to get theri parameters

3d mesh: objectes positions from lidar, color from the camera.

 

3 comment

After all the presentation, Dr. Li gave commnets on their project. chanllenge, achievement,,

report after presentation, due May 5.