Switching From Pixellot to Veo

Picture of a Veo2 from the Veo website

After a couple of seasons using the Pixellot camera, we made the switch to a Veo2 about eight games ago. In summary, yes, it’s better — but it’s not nearly as good as I thought it would be.

I’m a Football Analyst for an English tier 6 women’s football team — Steyning Town — who, alongside a Veo Camera, also use PlayerData as their GPS tracker.

Here’s a video I put together to show the players how to use the Veo Editor, which will give you an idea of how the platform works if you’re not familiar with it:

A run through of the editor I prepared for our players

One big caveat to the comparison below is that I don’t mention price or product tiers. I’ve been fortunate enough just to be given this tech to use, and it may well be that the downsides I mention below are resolved by throwing more money at either of these solutions. Clearly cost is a consideration though, and does lead into is one of my overarching hypotheses for the thought I’ve been giving to use of AI cameras and GPS trackers in grassroots sports — and that is that coaches don’t necessarily have the time and the skills to make best use of the tech that their clubs invest in. In terms of how to improve that, the focus of my work with Steyning has been to provide both my coaching colleagues and our players with a single place (a team and player performance portal) in which they can easily consume the data, video and any associated insights. Following on from this, two other areas I want to start to give some time to are:

1. If, as a club, you only have budget for an AI camera or GPS trackers, but not both, what do you invest in?
2. What ways are there in which clubs can get a return on that investment, so it starts to pay for itself?

But I digress….

What’s Good About Veo than Pixellot?

I’ve broken this down into things that are better than Pixellot — and then things about the Veo I really like.

What’s Better Than Pixellot

The four biggest obvious upsides to the Veo over the Pixellot are:

1. The picture quality is far better.
2. The match events that the camera detects are far broader — Pixellot just picks up shots, goals and restarts, where as Veo additionally picks up free kicks, penalties, corners and goal kicks
3. The zoom/panoramic/key frame capabilities are really good for more detailed analysis and better highlights. On the Pixellot these features were pretty much unusable — as performance in the editor was way too slow (see video below).
4. Telestration from within the Veo editor — although frustratingly you can’t export these, so if you want to share your telestrations and any associated commentary, we have to do so by using something like Quicktime to do a video screengrab we can narrate over.

Here’s a video comparing the panoramic features on Veo and Pixellot:

What Else is Good in General

1. I like the fact that Veo are very open to hearing about ideas for product improvements — although there’s quite a few heavily upvoted ideas which have been on the list for quite a while.
2. They actively involve their customers in the testing of new features.

What’s Not as Good About Veo

I’ve broken this down into things that aren’t as good as Pixellot, and things which just aren’t that good.

Things That are Better on Pixellot

1. Generating highlights is a real pain on Veo. On the Pixellot it’s much easier to select certain highlights and download them to create highlights reels elsewhere. I think improving this could be a really quick win for Veo, even if those highlights packages were only viewable within their walled garden. For example they could extend the functionality of their Live mobile app to make it easier to view previous games and highlights — not just those that have been streamed live. Given that the camera is expensive, the sponsorship we get for highlights packages is one of the key income streams that funds our investment, so it’s pretty frustrating that quality time that we could spend actually analysing video is spent knocking up highlights.
2. Pixellot’s processed clips give an indication of what players are involved by shirt number and then carries this analysis into the stats. Veo doesn’t do this.

Things That Just Aren’t so Good

1. While the match events that the camera picks up are far broader than Pixellot, it’s nowhere near 100% accurate and there are few options to effectively be a level of quality control over the top of this. For example, too often the AI misses goals — this is an easy fix as its possible to just create a custom event. But let’s say a goal is attributed to the wrong shot — i.e the corner taker and not the player who heads it in — there’s no fix for that. This means that we can’t rely on the data to be 100% accurate. Computer Vision automation is just not at a level yet where the models can work without human intervention (which would also additionally help train the models).
2. Match events that the AI picks up also can’t be augmented — so if you want to tag a player in a clip of, say a corner, you have to create a near duplicate event.
3. Coaches are unable to see player profiles. This is mega frustrating as all it’d take to improve this is some config around user roles and permissions. This is a missed opportunity, because player profiles would be a really good space for coaches and players to have individual conversations around performance. All we can do is tag players into clips (notwithstanding the challenges of my previous point) so that they go into their profiles, but there is no easy way of actually sitting down and adding a commentary around multiple clips that you might have tagged a player on across a number of games.
4. In a similar vein, it’s not possible to search across a range of matches for a particular event type. So for example, if we wanted to analyse how we defended corners across four or five games we need to go into each match individually.
5. Their customer care team believes too much in the product. Great I guess, but I spent 45 mins on a support chat outlining a bug, which the support rep insisted was a feature. It was a bug, that was later fixed, but his suggestion on the call was that what I was reporting was a great idea for a feature and that I should add it to the ideas list!

For some of the shortcomings above, we’ve managed to find a few workarounds:

1. For the player profiles we are continuing to use those that we built as part of our team and performance portal (which aggregates GPS tracking data alongside video and other performance insights).
2. In terms of being able to look at certain set pieces across range of games, I have imported all the individual clips into a separate system which I’m able to search much more effectively, I can output the search results into our team/player portal as well.

Player Telemetry

One of the features that I thought was going to be really useful was the ‘moving dots’ — what Veo call the 2D radar.

While we get a really accurate picture of how our players move on the pitch using the GPS data that we get from PlayerData, the shortcoming of that is that we only ever see data that is generated by our own players, so we have no idea of how our players move inline with the opposition. It’s more difficult to break down some of that positional data that we get with PlayerData into key game segments. For example, when we went a goal up did we start defending too deep?

In theory, Veo fixes both of these issues. I can see the moving dots for both teams and I can break it down by segments. In practice? It’s as good as what you’d expect a single fixed camera to be using computer vision to map the movement of two individual teams across a 3d environment, where it has to deal with issues like occlusion, match officials, coaches, subs warming up etc.

Here’s a video of a test I did of the feature using a goal we scored and a corner:

A test of the 2D radar ‘moving dots’

The holy grail here would be a system that could combine by video and GPS data, so I was really excited when I found out about Fitogether’s wearable and optical hybrid tracking, although this doesn’t seem to be commercially available at the moment.

Digital decisions are never a walk in the park, so please get in touch and let me help you find the right way through the technical landscape.