Why we stabilise instead of crop

The obvious solution nobody mentions

When you record video on a phone mounted to a dashboard, the footage vibrates. Wind buffets the device. Road texture shakes it. Engine idle makes it dance. Every smartphone video app faces this problem, and the industry standard answer is simple: crop the frame.

You record at 1080p. Shift the image in software during playback to counteract motion. You lose the edges, sure, but the centre stays still. Your video looks professional. Your app ships on time. Everyone does it this way.

We almost did too. Cropping is fast, computationally cheap, and it works. But then that Manchester clip landed in our inbox, and something clicked. If you crop away the edges, you crop away evidence. A dashcam is supposed to see what happened. If an insurance adjuster or a police officer watches your stabilised footage and sees less of the scene than was actually there, you've weakened your own case without realising it.

We'd been building Hawk to be court-ready. Integrity hashes on every clip. GPS overlay. One-tap export with a manifest so there's no ambiguity about what you recorded and when. Why would we then deliberately shrink the field of view to make the software easier?

Optical flow does the work instead of the knife

Optical flow is the harder path. Instead of cropping, it estimates the motion of pixels between frames and shifts the image to compensate. It's like tracking a pattern through a video and undoing the shake at its source, rather than just cutting the sides off.

The mathematics is more expensive. It takes more processing power. And on a phone, processing power is precious. Battery, CPU, memory. All of it costs money and heat and speed.

But the result is different. Your full frame stays intact. Everything that was in front of the lens stays in the footage. When you export that clip as evidence, you're sending the complete picture. Not a cropped, apologetic version of it.

We went down a rabbit hole with this. Early prototypes were sluggish. We tested on iPhone SE, older Android phones, devices with modest chipsets. We had to optimise the algorithm until it ran smoothly enough that you could record a whole shift without draining your battery by noon. The team spent weeks on it. It felt excessive at the time.

Then a Rideshare Pro subscriber sent us a message. She'd been in a minor collision. The police asked for footage. She sent the clip directly from Hawk, GPS overlay and all. The officer called back to say he could see the number plate of the other car, the traffic light colour, even the street sign in the background. All things that would have been cropped out of the frame with a simpler method. The case closed within a week.

Why this matters when stakes are real

A dashcam is not art. You're not recording it for YouTube. You're recording it because someday, someone might ask you to prove what happened. Insurance companies, police, maybe a judge if it goes that far. In those moments, your footage either helps you or it doesn't. There's no middle ground.

New drivers especially feel this pressure. Someone's texting you about a dent they swear you caused. Your insurance company wants evidence. You're stressed because you actually weren't at fault, but you need to convince someone else of that using a video clip.

Commuters face it too. A junction you use every day. One day someone runs a light and hits you. Your Hawk footage is the only independent record of what the traffic light actually showed. If that evidence is cropped down to hide what's at the edge of the frame, you've sabotaged your own case.

We chose optical flow because Hawk is supposed to be the thing you trust when it matters. Not when it's convenient or cheap or easy to ship. When it actually matters.

The trade-off we're okay with

Optical flow isn't perfect. On very extreme motion or sudden jerks, the algorithm can occasionally produce artefacts. A few pixels might shimmer or double up. It's rare, and it's never enough to damage the integrity of what you're recording. But it exists. We could have hidden it, pretended the technique was flawless.

Instead, we documented it and left it as is. Because the alternative of cropping away evidence is worse. And because building something honest sometimes means admitting where the seams show.

The battery hit is real too. Optical flow processing uses maybe 5 to 8 percent more power than a cropping approach would. On a full shift with the cabin camera and shift mode running (Rideshare Pro features), that adds up. We've optimised it as much as we can. But we won't pretend the trade-off doesn't exist.

What we will say is this: if you're recording evidence, you want the full picture. You want the edges. You want the context. And you want the person watching that clip to see exactly what your phone saw, without any frames cut away.

The question we ask ourselves

Every time we update the stabilisation, we still think about that Manchester driver. The clip that looked like bad driving but wasn't. The moment it became clear that every technical choice in Hawk should serve evidence, not convenience.

We're not the first app to use optical flow. We won't be the last. But we're probably the only one thinking about it in terms of court testimony and insurance claims and what happens when your video is the deciding factor in whether someone believes you.

That changes everything about how you build it. You don't optimise for beauty or smoothness alone. You optimise for integrity. For completeness. For the person who's going to watch this clip when it matters most, and know they're seeing the truth.