head-inthe-cloud/lidar_based_cone_detection

Micro-AV Lidar

Set up Environment

• Open User Manual for YDLidar X4 on the YDL official website
• Go to Section 3 - Linux ROS Operation, and follow steps 3.2:
  • Download SDK driver
  • Compile and Install YDLidar-SDK $ git clone https://github.com/YDLIDAR/YDLidar-SDK.git $ cd YDLidar-SDK/build $ cmake .. $ make $ sudo make install
• Install Dependencies $ cd YDLidar-SDK $ pip install . or $ python setup.py build + $ python setup.py install $ pip install matplotlib
• Open YDLidar-SDK/python/examples/plot_tof_test.py and change line 26 to this: laser.setlidaropt(ydlidar.LidarPropSerialBaudrate, 128000)
• Run Test $ cd YDLidar-SDK/python/examples/ $ python plot_tof_test.py
• Your lidar should run and you can see a map on your screen
• Click the map and press 'q' to stop operation

Getting .csv as result

• Download tof_test.py from this repo and replace the file in YDLidar-SDK/python/examples/
• run python tof_test.py
• Data will be stored in the specified folder

Plotting results

• Use plot.py to plot results from csv files, don't forget to change change data path

Dataset

Data is collected outdoors with a YDLidar x4

Configurations for ./data/12052021:

• Range: 2m, 4m, 6m, 8m, 10m
• Cone Location: The cone should be located at 180 degree angle

Configurations for ./data/12202021:

• Range: 1m, 2m, 3m, 4m, 4.5m, 6m
• Cone Location: The cone should be located at 0 degree angle

For CSE 455

Objective

The objective of the LiDAR cone detection part of this repo is to develop a quick, robust algorithm to detect certain landmarks and their location (e.g. a traffic cone) in order to provide information for autonomous vehicle localization and mapping (SLAM).

Related Works

[DROW - Deep learning based 2D Range data object detection](https://arxiv.org/pdf/1603.02636.pdf)

A visualization of the workflow of the algorithm could be found on [Miro](https://miro.com/app/board/uXjVOWxLUZA=/?invite_link_id=217485385838)

Approach

The main supporting theory of the cone detection algorithm is the Bayes Theorem

In our case, the theorem looks like this:

Pr (location | data) = Pr (data | location) * Pr (location) / Pr (data)

Pr (data) can be treated as a normalizing constant

Please see the PowerPoint Presentation for details of the algorithm.

More details of the cone detection algorithm and SLAM algorithm can be found here

Current Results

The green points represent input data, the red points represents the predicted centers, the blue dot and circle represents the ideal cone location.

Video

Link