1 / 13

(Non Technical) Overview of Deep Learning Object Detection on 3D Lidar Points

A comprehensive overview of object detection using 3D Lidar points, including its basics, limitations of cameras, and the need for Lidar technology. Learn about data sets, training data transformation, Python implementation, and measurement of accuracy in this non-technical primer. Explore the Fully Convolutional Net structure, accuracy metrics, and techniques like DBSCAN clustering. Dive into Korea's team results in the Didi-Udacity Challenge and considerations for faster processing. Discover the importance of a bird's eye view for panoramic data and the need for speed in processing. Join the journey into the world of 3D Lidar object detection.

chatman
Download Presentation

(Non Technical) Overview of Deep Learning Object Detection on 3D Lidar Points

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. (Non Technical) Overview of Deep Learning Object Detection on 3D Lidar Points Han Bin Lee Seoul Robotics

  2. Prelude • How it all got started • Udacity • 2017 Didi-Udacity Challenge

  3. Object Detection overview • Limitation of Camera • Need for Lidar

  4. Basic Structure of Lidar • 4 by n • x y z R + timestamp

  5. Big Data Used • Kitti data set • Udacity data set • All labeled and annotated

  6. Training Data Transformation • 2D panoramic transformation • Any points within annotated box was the ‘car’

  7. Python Implementation of 2D transpose

  8. The Deep Learning Structure

  9. Fully Convolutional Net • Fully Convolutional Network

  10. Measurement of Accuracy • Intersection over union value

  11. Some other Technique used • DBSCAN Cluster method

  12. Team Korea’s Result • 10th out of 2000

  13. After Thought • Data datadatadata • Limit of panoramic view – perhaps birds eye view was better • Data datadatadata • Need to be fast ~ 20hz

More Related