Quick look at the vehicle passing data from my 3,757 mile bike ride over 3 weeks from Edmonton, Canada to Hoover, Alabama through 2 Canadian provinces, 12 US states, and 115 counties.
I did Tour Divide in 2022, and I definitely remember the traffic "wake-up call" of coming into Helena, Montana after so many miles off road and away from all traffic. For the lateral passing distances, I've got an iPhone Pro 12 mounted sideways on the handlebar and have written an app to use the built-in LiDAR in iPhone Pro models (12 and later) to capture a depth map which tells you how far away everything is in the picture. I'm using an AI model to identify which part of the picture has the car, and then scanning the pixels within the boundary of the car to determine the closest point. The picture at the bottom of this post, shows how the AI model has identified the car (red), and then my algorithm has used the depth map to determine the closest point.
I am working this summer to train up my own AI model to use the rear video frames coming from the Varia RCT 715 to determine lateral passing distance. The idea is to use the LiDAR data as ground truth and basically say if the video looks like this, then the car passes at 1.0m (for example). With enough video and LiDAR data, I'm hoping to build a system that is accurate enough to at least get good approximations of how close cars are passing and build up a heatmap showing the difference in passing distances from people contributing videos from around the world.
I took on something similar last year (Tour Divide), and while my interactions with traffic were lesser, given the focus on off-roads, fire roads, etc., the route does present some challenges with vehicle traffic, and even a few fatal collisions over the last few years.
I definitely believe there is a 'local driving culture' that influences vehicle behavior in relation to bike traffic, and that is extremely inconsistent place-to-place, and also dependent on road types.
It's one of the things that is more frightening about on-road tours, to me, and a large part of why I favor dirt roads and the like.
I anxiously await the lateral passing distance data.
I'm curious (but willing to wait for that article, as I expect you are busy) if this lateral passing distance is something natively captured by the radar taillights, or if you had a secondary piece of tech for that?
It is indeed a secondary piece of tech ... I've got a sideways mounted iphone that uses lidar to build a depthmap of the car as it passes. I am currently working with one of my students to train up a neural network to pair this lidar data with the rear-facing video camera to predict the passing distances based on the approaching video. Eventually I hope to roll this out to everyone so that anyone with a rear-facing camera will be able to contribute to the lateral passing distances data.
Yes, I got mine a couple days ago. It is amazing! But they are hiding the new features behind secure bluetooth and not exposing the data to Garmin ConnectIQ developers. So for now, I'm going to keep running my RCT715, since it also records video, whereas the new 820 does not. Also, without the proper Garmin firmware credentials, third parties will not have access to the new features either. I have contacted Garmin in several different ways, and am hopeful that they change their mind about exposing the lateral data to Connect IQ developers, which is literally only compatible with Garmin products.
Bummer that that data isn't natively open for developers.
I hope your efforts with them pay off.
I just received mine, but have not yet set it up.
I'm excited about the battery life and USB-C, most of all, and the other features look like gravy on top.
I have some other ideas for how the new features on this device could help detect drivers that either swerve purposely at cyclists, or do so because they are inattentive, give a stronger alert to the cyclist, and also create an obnoxious flashing pattern to try to get the driver's attention (or make them think they might be being recorded).
Oh hey, I also tried to friend you on Facebook recently, if you are so inclined.
We seem to have several things in common besides the bike light interest.
I did Tour Divide in 2022, and I definitely remember the traffic "wake-up call" of coming into Helena, Montana after so many miles off road and away from all traffic. For the lateral passing distances, I've got an iPhone Pro 12 mounted sideways on the handlebar and have written an app to use the built-in LiDAR in iPhone Pro models (12 and later) to capture a depth map which tells you how far away everything is in the picture. I'm using an AI model to identify which part of the picture has the car, and then scanning the pixels within the boundary of the car to determine the closest point. The picture at the bottom of this post, shows how the AI model has identified the car (red), and then my algorithm has used the depth map to determine the closest point.
I am working this summer to train up my own AI model to use the rear video frames coming from the Varia RCT 715 to determine lateral passing distance. The idea is to use the LiDAR data as ground truth and basically say if the video looks like this, then the car passes at 1.0m (for example). With enough video and LiDAR data, I'm hoping to build a system that is accurate enough to at least get good approximations of how close cars are passing and build up a heatmap showing the difference in passing distances from people contributing videos from around the world.
Wow, thank you for the detailed response, and I am super impressed with your inventiveness and know-how.
You're inspiring me to rethink what i possible for data capture, myself.
Wow, that looks like an adventurous trip!
I took on something similar last year (Tour Divide), and while my interactions with traffic were lesser, given the focus on off-roads, fire roads, etc., the route does present some challenges with vehicle traffic, and even a few fatal collisions over the last few years.
I definitely believe there is a 'local driving culture' that influences vehicle behavior in relation to bike traffic, and that is extremely inconsistent place-to-place, and also dependent on road types.
It's one of the things that is more frightening about on-road tours, to me, and a large part of why I favor dirt roads and the like.
I anxiously await the lateral passing distance data.
I'm curious (but willing to wait for that article, as I expect you are busy) if this lateral passing distance is something natively captured by the radar taillights, or if you had a secondary piece of tech for that?
It is indeed a secondary piece of tech ... I've got a sideways mounted iphone that uses lidar to build a depthmap of the car as it passes. I am currently working with one of my students to train up a neural network to pair this lidar data with the rear-facing video camera to predict the passing distances based on the approaching video. Eventually I hope to roll this out to everyone so that anyone with a rear-facing camera will be able to contribute to the lateral passing distances data.
Hey Brian - did you see the new Garmin RearVue 820 that launched today?
Yes, I got mine a couple days ago. It is amazing! But they are hiding the new features behind secure bluetooth and not exposing the data to Garmin ConnectIQ developers. So for now, I'm going to keep running my RCT715, since it also records video, whereas the new 820 does not. Also, without the proper Garmin firmware credentials, third parties will not have access to the new features either. I have contacted Garmin in several different ways, and am hopeful that they change their mind about exposing the lateral data to Connect IQ developers, which is literally only compatible with Garmin products.
Bummer that that data isn't natively open for developers.
I hope your efforts with them pay off.
I just received mine, but have not yet set it up.
I'm excited about the battery life and USB-C, most of all, and the other features look like gravy on top.
I have some other ideas for how the new features on this device could help detect drivers that either swerve purposely at cyclists, or do so because they are inattentive, give a stronger alert to the cyclist, and also create an obnoxious flashing pattern to try to get the driver's attention (or make them think they might be being recorded).
Oh hey, I also tried to friend you on Facebook recently, if you are so inclined.
We seem to have several things in common besides the bike light interest.