sighthound improvements

[camera_kat]

The program saves all video and then lets you object search on the search screen, I'd rather have a mode where it only saves the video with the search parameter I set up on the camera screen- and nothing else.

On the Search screen, my search is on the very bottom of the "look for" list. This is terrible, it should be on the top, and would be solved by the request above.

Why are the region zones linked to my searches? I'd like to rope off an area so cars driving by don't get recorded all day long. It is really weird that boundary boxes are only inside search rules and not for motion detection elimination. Two 1080p cameras will chew right through 100 gigabytes of storage a day. This is completely ridiculous and unnecessary. In fact, I may have to just stop using sighthound until this is fixed.

The accuracy of the people algorithm. The program thinks headlights against a garage door are people. It thinks headlights sweeping across the lawn are people. Thankfully it thinks actual people are in fact people, otherwise I'd be really annoyed. I'd suggest a "this identification is incorrect" button that will optionally send a screen shot to you guys for algorithm improvement.

When I set up my rule, the area of the screen for moving the boundary box around is REALLY SMALL. Why cant you have much larger, much more accurate screen for setting up the boundary box?

Why in the name of all that is holy must this program remain running on the console? Could you please make it run as a windows server service? Please? Seriously, pretty please?

HTML5 please. Why make me install Shockwave/Flash for the web UI? It's the work of the devil!!!

Force TCP! Every time a UDP packet gets lost in the ether, there is a frame glitch, which results in a motion triggered recording event.... Ugh.. This happens a LOT.

The video camera frame rate is 30 frames per second, yet the sighthound exported or saved clips are 10 frames per second. Why does it do this?

Thanks for listening...

[Semper Vaporo]

In reality, it does not matter how much video is saved to the disk, the only data that ultimately is kept is that which matches a rule (and that is pared down to just that segment). The rest of the video data, where the motion does not match a rule or where there is no motion at all, is deleted after a user determined time period. Even that which is saved by a rule will eventually get deleted when you run out of disk space... but the first data to go is the oldest "not saved by rule" data, followed by the oldest "saved by rule".

By default the raw video is kept for 24 hours, but you can change that to any number of hours, using the "TOOLS|Options" menu and selecting the "Storage" tab.


Each rule can have different bounding boxes so that is why they are tied to the rules instead of the overall scene.


As for the "Look in:" list... I can understand the desire to have a User's rule "on top" of the list, but I have 6 cameras and 8 rules total, so which one goes "on top"? My preference would be for that list to be option buttons for the predefined searches and a 4th option to select the rule from a list box that only lists the user defined rules. And that list box needs to be wider to show more of the name of the rule!

I'd also like the "Camera" list to be a scrolling list, with several names visible, instead of a drop down.

I am sure that the SightHound people will take your comments into consideration in future enhancements and updates.


Motion detection is relatively easy, but determining which is a "person" and which is not, is something that is very difficult to do. A simplistic version is to detect the relative height to width, but SightHound goes further than that to try to decide if that motion is from a person. Click on the word "Technology" in the banner at the top of any SightHound page and follow some of the links in the web page that appears. It is not a simple process that SightHound follows!

[camera_kat]

In reality, it does not matter how much video is saved to the disk

It TOTALLY matters how much video is saved to disk. SSD's have limited write endurance. When Sighthound is recording cars driving by 24/7 outside of a boundary box, then it is literally wasting money by wearing out the SSD.

[Semper Vaporo]

Well... you didn't say that you were writing to a limited usage drive, so that might be a consideration.

I have been considering getting an SSD to see how that might improve my system, but this "limited write lifetime" has been a concern to me and has slowed my action in that regard (not to mention the $$ involved!).

But I wonder... from what I have read about SSD drives, that limit to the number of write cycles it can endure is somewhere in a range above 100,000 writes per memory location (i.e.: per bit). That appears to be small in the scheme of things, but just fer instance... if you have a 100 GByte drive and you are writing 100 GBytes per day, then you have 100,000 days of write cycles. Is that not about 273 years before it will wear out? That sure seems to be a non-worrisome problem... But considering all the industry and media concern with it, what is wrong with my math here?

[camera_kat]

Well... you didn't say that you were writing to a limited usage drive, so that might be a consideration.

I have been considering getting an SSD to see how that might improve my system, but this "limited write lifetime" has been a concern to me and has slowed my action in that regard (not to mention the $$ involved!).

But I wonder... from what I have read about SSD drives, that limit to the number of write cycles it can endure is somewhere in a range above 100,000 writes per memory location (i.e.: per bit). That appears to be small in the scheme of things, but just fer instance... if you have a 100 GByte drive and you are writing 100 GBytes per day, then you have 100,000 days of write cycles. Is that not about 273 years before it will wear out? That sure seems to be a non-worrisome problem... But considering all the industry and media concern with it, what is wrong with my math here?

That is assuming the OS and none of the apps write anything. That is assuming it is being used on a single user workstation and not a VM host. Most people are going to want to run sighthound in a virtualized environment so as not to dedicate an entire machine to host it. That is also assuming one wont use more than two cameras or that even higher resolution cameras are not on the horizon.

[Walt Dockery]

what do you mean most people will want to run in a virtualized environment?

[camera_kat]

what do you mean most people will want to run in a virtualized environment?

Most people with a decent amount of cameras are going to run this in a VM. They aren't going to waste an entire PC running this.

[Semper Vaporo]

A VM is a "Virtual Machine"... it uses CPU time to run the program in the "VM" as well as CPU time to handle the interface between the real hardware and the 'pretend' hardware that the program running in the VM thinks it is accessing.

Thus, running any program in a VM is slightly more CPU intensive than not doing so.

I only know the Winders world, so I can't speak to what happens in the MAC world, but Winders is a multi-processing and multi-programming system capable of running multiple programs pseudo-simultaneously. But even if one of those programs is in a VM, it is still using the same CPU(s) as the VM machine and the VM machine is sharing that/those CPU(s) the same as running any program.

I know that I can run SightHound and then minimize the User Interface window or terminate the User Interface program and the sub-processes continue running, saving data to the disk, and detecting motion "on the fly" and issuing the various outputs as specified by the user's rules. I don't need to waste CPU cycles acting as in interface between the VM and the real world. And while SightHound is running (even with the User Interface visible, or minimized or terminated), I can do other things... but the more cameras there are and the higher resolutions those cameras are and the more frames per second the cameras are sending, the less CPU time there is available to do those other things.

I do use a VM when I want to run something of unknown background so that if it does something that might damage my system (such as modifying a system file) I can kill the VM and nothing in the "real" machine is modified.

[tesla8er]

@camera_kat,

Thanks for the comments. In reply:

The software records all video so that you have the option of gong back and seeing things you missed, or to look at what happened between clips that were captured. It's a feature people like. If you don't want to take up space on your hard drive to do that set the Temporary Video setting to 1 hour. (The default is 48 hours)

The bounding boxes are specific to the rules. If the concern is using drive space to record unwanted scenes, lower the temp video setting and use the 'ignore objects outside this area' setting.

We'll look at the ordering of the lists. I don't recall that being requested before, perhaps because of Semper's comment, that by the time you have lots of rules you'll have to scroll anyway.

The algorithms for detecting people and objects are improving all the time. It's a hard task and we're not perfect, but each iteration of the computer vision algorithms reduces the false positives.

We haven't been asked for a larger screen for drawing bounding boxes before. It seems large enough to draw lines around doors and mask off roads etc.

It will run as a service soon. We should have done that from the start. Note that if the cameras are on and you shut the console SV keeps recording as long as you prevent the computer going to sleep.

The web UI will get a makeover.

You can force TCP yourself. See other threads for the command. We're looking at the default discovery and default setting for this to save people the hassle of adding the command if their network is losing packets.

The optimum rate for different computer vision tasks is not necessarily 30 fps. Optimizing stream for viewing and a stream for analyzing is one of the ideas under consideration.

Only a tiny fraction of users report using SV on a VM. Given you can reduce the temp video kept, control the volume of clips saved, close the console, run SV in the background and turn off the monitor, most people just let it run. The peak CPU usage is when SV is doing disk clean up or sorting clips, often at start up. Mac users hide Mac minis in closets and just leave it running.

Overall, there's a lot of development work going on and we do listen to user requests. You'll see various improvements arrive in the coming months.

Stephen
Sighthound

[camera_kat]

use the 'ignore objects outside this area' setting.

Where is this setting? I cannot find it. Thanks.