During these two weeks, I continue to revise the paper that I want to conclude the former works. Basicly, I have finished. But I still found many place need to be refined again. I have upload it to gitlab.

This week I am still writing that paper. The paper is already finished, but still find many places to modification.

For better analysis and comparing of different systems, I did more experiments. Also for visualization figures in my paper, I also adjust my proposed work.

1. Paper reading: "yolov4-5d", This paper did a very good survey on autonomous driving, and I can learn a lot about how to write "introduction" part in my later paper.

2. About ROS, I did the basic learning and experiments through ROS, but I still not use ROS in … Continue Reading ››

1.AROB: I have modified the conference paper, and submitted the revised paper for AROB Journal.

2.ROS: I have acquired the basic skills of AROB, and try to make the point cloud data for visualization through rviz tools.

I spend some time to learnt the ROS system, and used my laptop to connect the LiDAR sensor. I plan to use LiDAR only at first, not add camera sensors.

I tested the LiDAR sensor yesterday, and found no problems for construct point cloud data.

From the paper I read before: Frustum-pilliars, I go further detailed study of 3D point cloud encode method: point-pilliars.

In this paper, I found the encode method can be divided into two parts: pesudo-images and 3D conv.

pesudo-images includes: complex-yolo and point-pilliars

3D … Continue Reading ››

This week, I continue to did some deep-learning framework construction. I learned from the open source frameworks.

This week, I mainly learnt the detailed programming work of the open source framework, and try to do it myself. The programs are YOLOP and Complex-yolo.

1. I searched the lidar-camera fusion calibration method. Matlab and ros both have the calibration bag.
2. With the calibration data, the 3D lidar points can be projected to 2D images. So I read a mutil-stage 3D detection paper:"Frustum-PointPillars: A Multi-Stage Approach for 3D Object Detection using RGB Camera and LiDAR". This work used … Continue Reading ››
   

1. I prepared for the AROB presentation and attended the meeting.
2. When I get the LiDAR, I need to do calibration work with cameras. I am now searching for some usable methods for this work.