Traffic Light Tracking
March '18
Traffic Light Tracking
March '18
Computer Vision
March 2018
Background
LiDAR- what it is, why it is relevant
Reference Velodyne LiDAR page
Remote Sensing- class at UMD, build our own LiDAR systems to map something on campus --> note less design, more build
​
Build Process
- explanation of layout/how it works
- electrical components and general layout provided
-material selection
- actual build/assembly process (video)
​
LiDAR System
February - May 2018
Background
LiDAR (light distance and ranging) is one of the fundamental technologies used in autonomous vehicles. It disperses light through a series of lasers and uses the reflected light to determine distance.
​
This project was completed for a Remote Sensing course I took in my final semester at Maryland. We were tasked with building our own basic LiDAR systems and using those to map something on campus.
​
Now, the LiDARS frequently used on self driving cars have an array of lasers and receivers that spin multiple times per second. In this project, we created a geometric LiDAR that doesn't spin, but uses known geometry combined with computer vision to determine the distance to an object.
​
I opted to map a ~40ft clock tower on Maryland's campus. My main goal was to map the base of the clock tower, but I hoped to be able to collect sufficient date to get about half the height of the tower.
At the time, I didn't fully understand the limits of the hardware I was working with before deciding to map this object. As a result, I encountered numerous hardware related challenges which lead to some very valuable lessons.
Hardware
_edited.jpg)
Raspberry PI
Arduino
Line laser
Camera
12''
For this project I used a Raspberry Pi connected to a Raspberry Pi camera, an Arduino, and a line laser. A cellphone battery pack was used to provide power, but unfortunately is not pictured.
The Arduino in this case is used strictly to provide power to the line laser while the Raspberry Pi captures and stores images during the data collection in preparation for post processing.
​
The mount is made out of plastic and was 3D printed on an FDM printer. Finally, the mount is attached to a piece of wood which ultimately attaches to a tripod. The full configuration is oriented vertically.
Vertical configuration on tripod
How a Geometric LiDAR works
Laser line
Flat object
Top view
_edited.jpg)
Say we moved a flat object in a straight line from right to left towards the LiDAR.
Flat object
Laser dot
The Camera's Perspective
Recall that the geometry of thee system and the location of camera/laser are fixed with a 12" separation.
Therefore, as the flat object moves in a straight line closer to the LiDAR, the camera sees that the laser dot moves rightward.
Using some computer vision to detect the moving laser and using the known distance between the laser and camera, the distance from the LiDAR to an object can be calculated. As a result, something can be mapped.
Calibration & Test Process
Before using the sensor to collect data, I had to make sure it aligned with real world values.
I collected a screenshot of the laser lines on the Raspberry Pi at six inch increments for 10ft. I then ran a calibration script to plot how closely the sensor matched theoretical data.
Finally, I made minor adjust mentions to angular values to bring the alignment closer to the theoretical.
To ensure the collection was working properly, I mapped a simple piece of plywood indoors.
​
Mapping in the Field
Mapping of the clock tower required a few prerequisites. First, the night had to be completely clear. Since the Raspberry Pi, laser, and Arduino were not weatherproofed, data could only be collected on a clear night.
​
Furthermore, the Raspberry Pi had to be accessible via laptop through vpn. This way I could see the quality of the laser line before collecting the data.
​
Finally, and perhaps most importantly, the clock tower had to be vacant.