Hello, this is for the staff of Sighthound. I am comparing various video surveillance systems, and would like to know if SightHound has the following features:
Size Detection
Color Detection
Speed Detection
Distance Detection
Lighting / Shadows Detection
Repeated Motion Detection
Rain Detection
Wind Detection (eg. leaves)
Object Detection (eg. car)
Animal Detection (eg. dogs, squirrels, etc)
Human Detection (vs animals, objects)
Facial Recognition
Night Detection - how good is it?
Zones of interest
Movement Type - running, loitering, fighting? elaborate
Machine Learning - intelligent learning algorithms? elaborate
Searching / Tags
Rules / Triggers
Alerts / Notifications
Cloud upload and storage
Security - what types of security features?
Bandwidth usage - will it slow down my network?
SightHound features list comparison?
Moderator: Staff
6 posts
• Page 1 of 1
-
tesla8er - Posts: 30
- Joined: Sat Jun 28, 2014 3:17 pm
Re: SightHound features list comparison?
Hello Ron Burgundy,
Sighthound Video differentiates between people and objects. See https://www.sighthound.com/surveillance
In the support section you'll find how to set rules:
https://www.sighthound.com/support/rule-editor
You can use Remote Access to set mobile alerts. https://www.sighthound.com/support/topic/remote-access
We don't run our own cloud service. You can export to your account at Dropbox, Google Drive or others, here: https://www.sighthound.com/support/export-local-folder
Network bandwidth isn't normally an issue. https://www.sighthound.com/support/system-requirements
Above all, you can go here and get a free trial to see if it does what you want it to do: https://www.sighthound.com/download
Stay Classy Ron,
Stephen
Sighthound
Sighthound Video differentiates between people and objects. See https://www.sighthound.com/surveillance
In the support section you'll find how to set rules:
https://www.sighthound.com/support/rule-editor
You can use Remote Access to set mobile alerts. https://www.sighthound.com/support/topic/remote-access
We don't run our own cloud service. You can export to your account at Dropbox, Google Drive or others, here: https://www.sighthound.com/support/export-local-folder
Network bandwidth isn't normally an issue. https://www.sighthound.com/support/system-requirements
Above all, you can go here and get a free trial to see if it does what you want it to do: https://www.sighthound.com/download
Stay Classy Ron,
Stephen
Sighthound
-
Semper Vaporo - Posts: 331
- Joined: Sat Oct 09, 2010 11:51 am
Re: SightHound features list comparison?
As a satisfied user of SightHound let me take your points one at a time and comment on them.
Size Detection
The program does have the ability to set a lower limit of the size of an “area” that is “moving”. It is set as a number of Pixels in height. This works well in scenes where the depth of the view is limited and where there is little chance of something getting close to the camera, such as in an indoor room, with the camera high enough that someone cannot stand directly in front of it. Outdoor scenes can be a problem due to perspective… and elephant across the street can effectively hide behind a house fly on the lens of the camera.
Color Detection
Nope. Motion is motion, regardless of the colors involved
Speed Detection
Nope. If you know the distance between two points in the scene and have a stopwatch you can calculate the speed, but the program has no provision for either of those.
Distance Detection
Yes, to a certain extent, though maybe not what you mean. If the camera is mounted such that the scene has distant things near the top, and close things near the bottom then any movement that is detected where the bottom of the movement is near the bottom of the scene, then it can be inferred to be close to the camera and things where the bottom of the movement is near the top of the scene then it is probably further away. You can create a rule for motion detection where you set an area in the scene that detects movement only when the bottom of the movement is in an area near the bottom of the scene. This would exclude detecting motion in the distance. But the program will not tell you how far away an object is, nor does it have a setting that says only detect something if it is closer than xx ft.
Lighting / Shadows Detection
Nope… It detects “motion” regardless of the lightness or darkness of the scene and cannot detect that the movement is only a shadow and not an object that is just lighter or darker than the surrounding parts of the image. Shadows of trees swaying in the wind are no different than a person walking through the same area of the scene. Some shadow movement can be ignored by creating a rule that only detects movement that is “entering or exiting” a scene or area. Thus, if the movement is wholly contained in the area then it might not trigger the rule… this does not work well if the swaying tree shadow is near the edge of the scene or area.
Repeated Motion Detection
If you mean you want to ignore the swinging pendulum of a clock then, no, it cannot do so. But you can set your detection rules to exclude that area of a scene and thus not generate alerts because the pendulum is moving… but since Sighthound records ALL video ALL the time and records the time of ANY movement in a database, there will be some additional drive activity doing constant database updates and then removing the recordings after the user’s “keep all video for xx hours” setting. This can make for a full list of useless entries in the Review mode until that number of hours has elapsed. But you can set the Review mode to show only movement recorded “by rule”, and that will reduce the list to just that movement outside from the swinging pendulum area.
Rain Detection
Yep! Rain running down the window glass will be detected as motion, but I bet that is not what you are asking, so, no, it does not know when it is raining other than wind-blown sheets of rain might be detected, but only as “motion” in the scene, not as ‘Rain’.
Wind Detection (eg. leaves)
Again, yes, it sees “Leaves” blowing in the wind and does not differentiate that from a dog or human wandering around in the scene.
Object Detection (eg. car)
Animal Detection (eg. dogs, squirrels, etc)
Human Detection (vs animals, objects)
The program does differentiate between “Human” and “other objects”, but bases that on the vertical vs horizontal sizes. In general, tall thin moving areas are determined to be human and everything else is “other object”. This works well most of the time, but depending on the camera orientation it can be fooled. If the camera is viewing the scene at an angle of, say around 45-degrees down, then a Dachshund (wiener dog?) coming toward the camera might be seen as tall and thin and called a human and a person carrying a big box across the scene might be seen as wider than tall and be called an “other object”.
Facial Recognition
No.
Night Detection - how good is it?
This depends more on how good the camera is at providing an image to work with. I use mainly cheap Web-cams and they do not work well in poor lighting. But I have one that has a view out my front window toward a streetlight (just out of view) and the pool of light on the street is just barely visible… yet if it weren’t for the bounding boxes the program can draw around motion areas in the Review mode, I would be wondering at what it saw when I hear an alert at night. I have to go to the Review mode to determine what triggered it (usually a neighbor out walking his dog at 11:45 PM causing just a dark silhouette moving across the scene).
Zones of interest
You can set rules where you define areas where motion must be for the rule to trigger an action. There are many ways of utilizing this feature and it can be a very complicated shape to the area. I have effectively created an area that includes a whole room, except for the TV screen and the reflection of the TV image in a couple of shiny objects elsewhere in the room… makes for a ridiculous shape to the area, but works quite effectively!
Movement Type - running, loitering, fighting? Elaborate
Any movement is detected, but the reason for the movement is irrelevant to the program. There is a rule feature that might be useful for detecting “loitering” in that it will limit the rule trigger to only objects that are present for over a certain amount of time.
Machine Learning - intelligent learning algorithms? elaborate
Presently, the algorithms are built-in and not something where the end user can “characterize” individual scenes, but with the rules features a lot can be controlled as to what characterizes motion of interest and non-interest.
Searching / Tags
Don’t know what you mean by “Tags”. The ability to search through recorded videos from the cameras (until the limit of “keep all video for xx hours” set by the user) is quite well done. Beside the rules that initially flag video to be kept, additional rules can be created in the Review mode to help find other motion. Though, unfortunately, these new rules do not make the newly found video to be included in the more permanent saved video clips, like the ‘real-time’ rules do.
Rules / Triggers
As stated above, the program lets the user set many rules for detecting motion and limiting that detection to certain areas, times of day, size of object, center or bottom of the motion in the area, or crossing a line (in one or either direction), and many more limitations and refinements.
Alerts / Notifications
Yes, rules can be defined to play a sound on the computer speaker, or run an external program, or send an e-mail which includes an image from the scene.
Cloud upload and storage
Yes, it can upload video to internet storage.
Security - what types of security features?
Of its own, I don’t think there is any specific security feature provided, but it relies on the user’s anti-virus and firewall security features to limit access to the computer and the Sighthound program and files.
Bandwidth usage - will it slow down my network?
Depends on what the surveillance setup is composed of and what is happening at the time. LAN based IP cameras will use bandwidth and the more cameras the more bandwidth they will use. WiFi bandwidth is often about 1/3rd of a wired system so too many WiFi devices can be problematic. It also depends on the resolution and frame rate of the cameras and somewhat as to how busy with motion the scenes are composed of and the type of compression being used. Also, it may depend on how many alerts are being sent via the internet and if videos are being transferred to off-site storage. Then there are the other things that might be being done on the whole system… if your son is playing a real-time, 1st person, slaughter everything in sight game against a dozen other people via the internet, and your daughter is watching videos of the latest heartthrob band and your wife is streaming “Gone With The Wind”, all at the same time, they will be arguing with each other long before your IP cams are put on line… but adding them will add you to the argument. If you have multiple high-rez IP cams, I would recommend they all have their own router to keep the surveillance system out of the “mainstream”. If using WiFi for the cameras, set the separate router to a different channel (widely separated from the other WiFi channel). If you also will be using a connection to the internet that is more than just sending sparse Alert e-mails, you might consider a 2nd ISP modem as well.
Size Detection
The program does have the ability to set a lower limit of the size of an “area” that is “moving”. It is set as a number of Pixels in height. This works well in scenes where the depth of the view is limited and where there is little chance of something getting close to the camera, such as in an indoor room, with the camera high enough that someone cannot stand directly in front of it. Outdoor scenes can be a problem due to perspective… and elephant across the street can effectively hide behind a house fly on the lens of the camera.
Color Detection
Nope. Motion is motion, regardless of the colors involved
Speed Detection
Nope. If you know the distance between two points in the scene and have a stopwatch you can calculate the speed, but the program has no provision for either of those.
Distance Detection
Yes, to a certain extent, though maybe not what you mean. If the camera is mounted such that the scene has distant things near the top, and close things near the bottom then any movement that is detected where the bottom of the movement is near the bottom of the scene, then it can be inferred to be close to the camera and things where the bottom of the movement is near the top of the scene then it is probably further away. You can create a rule for motion detection where you set an area in the scene that detects movement only when the bottom of the movement is in an area near the bottom of the scene. This would exclude detecting motion in the distance. But the program will not tell you how far away an object is, nor does it have a setting that says only detect something if it is closer than xx ft.
Lighting / Shadows Detection
Nope… It detects “motion” regardless of the lightness or darkness of the scene and cannot detect that the movement is only a shadow and not an object that is just lighter or darker than the surrounding parts of the image. Shadows of trees swaying in the wind are no different than a person walking through the same area of the scene. Some shadow movement can be ignored by creating a rule that only detects movement that is “entering or exiting” a scene or area. Thus, if the movement is wholly contained in the area then it might not trigger the rule… this does not work well if the swaying tree shadow is near the edge of the scene or area.
Repeated Motion Detection
If you mean you want to ignore the swinging pendulum of a clock then, no, it cannot do so. But you can set your detection rules to exclude that area of a scene and thus not generate alerts because the pendulum is moving… but since Sighthound records ALL video ALL the time and records the time of ANY movement in a database, there will be some additional drive activity doing constant database updates and then removing the recordings after the user’s “keep all video for xx hours” setting. This can make for a full list of useless entries in the Review mode until that number of hours has elapsed. But you can set the Review mode to show only movement recorded “by rule”, and that will reduce the list to just that movement outside from the swinging pendulum area.
Rain Detection
Yep! Rain running down the window glass will be detected as motion, but I bet that is not what you are asking, so, no, it does not know when it is raining other than wind-blown sheets of rain might be detected, but only as “motion” in the scene, not as ‘Rain’.
Wind Detection (eg. leaves)
Again, yes, it sees “Leaves” blowing in the wind and does not differentiate that from a dog or human wandering around in the scene.
Object Detection (eg. car)
Animal Detection (eg. dogs, squirrels, etc)
Human Detection (vs animals, objects)
The program does differentiate between “Human” and “other objects”, but bases that on the vertical vs horizontal sizes. In general, tall thin moving areas are determined to be human and everything else is “other object”. This works well most of the time, but depending on the camera orientation it can be fooled. If the camera is viewing the scene at an angle of, say around 45-degrees down, then a Dachshund (wiener dog?) coming toward the camera might be seen as tall and thin and called a human and a person carrying a big box across the scene might be seen as wider than tall and be called an “other object”.
Facial Recognition
No.
Night Detection - how good is it?
This depends more on how good the camera is at providing an image to work with. I use mainly cheap Web-cams and they do not work well in poor lighting. But I have one that has a view out my front window toward a streetlight (just out of view) and the pool of light on the street is just barely visible… yet if it weren’t for the bounding boxes the program can draw around motion areas in the Review mode, I would be wondering at what it saw when I hear an alert at night. I have to go to the Review mode to determine what triggered it (usually a neighbor out walking his dog at 11:45 PM causing just a dark silhouette moving across the scene).
Zones of interest
You can set rules where you define areas where motion must be for the rule to trigger an action. There are many ways of utilizing this feature and it can be a very complicated shape to the area. I have effectively created an area that includes a whole room, except for the TV screen and the reflection of the TV image in a couple of shiny objects elsewhere in the room… makes for a ridiculous shape to the area, but works quite effectively!
Movement Type - running, loitering, fighting? Elaborate
Any movement is detected, but the reason for the movement is irrelevant to the program. There is a rule feature that might be useful for detecting “loitering” in that it will limit the rule trigger to only objects that are present for over a certain amount of time.
Machine Learning - intelligent learning algorithms? elaborate
Presently, the algorithms are built-in and not something where the end user can “characterize” individual scenes, but with the rules features a lot can be controlled as to what characterizes motion of interest and non-interest.
Searching / Tags
Don’t know what you mean by “Tags”. The ability to search through recorded videos from the cameras (until the limit of “keep all video for xx hours” set by the user) is quite well done. Beside the rules that initially flag video to be kept, additional rules can be created in the Review mode to help find other motion. Though, unfortunately, these new rules do not make the newly found video to be included in the more permanent saved video clips, like the ‘real-time’ rules do.
Rules / Triggers
As stated above, the program lets the user set many rules for detecting motion and limiting that detection to certain areas, times of day, size of object, center or bottom of the motion in the area, or crossing a line (in one or either direction), and many more limitations and refinements.
Alerts / Notifications
Yes, rules can be defined to play a sound on the computer speaker, or run an external program, or send an e-mail which includes an image from the scene.
Cloud upload and storage
Yes, it can upload video to internet storage.
Security - what types of security features?
Of its own, I don’t think there is any specific security feature provided, but it relies on the user’s anti-virus and firewall security features to limit access to the computer and the Sighthound program and files.
Bandwidth usage - will it slow down my network?
Depends on what the surveillance setup is composed of and what is happening at the time. LAN based IP cameras will use bandwidth and the more cameras the more bandwidth they will use. WiFi bandwidth is often about 1/3rd of a wired system so too many WiFi devices can be problematic. It also depends on the resolution and frame rate of the cameras and somewhat as to how busy with motion the scenes are composed of and the type of compression being used. Also, it may depend on how many alerts are being sent via the internet and if videos are being transferred to off-site storage. Then there are the other things that might be being done on the whole system… if your son is playing a real-time, 1st person, slaughter everything in sight game against a dozen other people via the internet, and your daughter is watching videos of the latest heartthrob band and your wife is streaming “Gone With The Wind”, all at the same time, they will be arguing with each other long before your IP cams are put on line… but adding them will add you to the argument. If you have multiple high-rez IP cams, I would recommend they all have their own router to keep the surveillance system out of the “mainstream”. If using WiFi for the cameras, set the separate router to a different channel (widely separated from the other WiFi channel). If you also will be using a connection to the internet that is more than just sending sparse Alert e-mails, you might consider a 2nd ISP modem as well.
Semper Vaporo,
-
tesla8er - Posts: 30
- Joined: Sat Jun 28, 2014 3:17 pm
Re: SightHound features list comparison?
Some clarifications on Semper's post.
Sighthound doesn't look at vertical vs horizontal boxes to guess if something is a human. It uses a neural network to assess a score based on body parts, eyes, head, etc being arranged and moving in the same way as the humans in videos it was trained with move. Pretty much the whole point of Sighthound is it doesn't use simplistic motion detectors or boxes to try to define motion. Here is a white paper explaining the technology. https://www.sighthound.com/files/Sighth ... r_2013.pdf
Color (or gray scale pixels at night) is used to track humans around the frame, but you can't search a video for a color in it.
The default setting for each camera added is to save clips whenever Any Object appears in the frame. That could be a leaf, a shadow, rain, a dog, a car, anything. Sighthound won't distinguish between them, it will just say Object Seen. If you change the setting to People, it will try to find only the objects that look very similar to the humans it was trained on. Does it get it right all the time? No. But it is successful enough that people don't get alerts from leaves, rain or dogs most of the time.
On Night Detection - if you have a decent IR camera, Sighthound will be able to see the scene almost as well as during the day. The tracking still works in gray scale, and the people detector works well.
On bandwidth we tend to see the video encoding and decoding slowing down the CPU before we see network issues. I'm sure if you add enough Hi Def cameras over Wifi and there is other stuff going on in the household, you'll clog your Wifi, however we have users with 25 cameras running on a big server with no network issues.
Stephen
Sighthound doesn't look at vertical vs horizontal boxes to guess if something is a human. It uses a neural network to assess a score based on body parts, eyes, head, etc being arranged and moving in the same way as the humans in videos it was trained with move. Pretty much the whole point of Sighthound is it doesn't use simplistic motion detectors or boxes to try to define motion. Here is a white paper explaining the technology. https://www.sighthound.com/files/Sighth ... r_2013.pdf
Color (or gray scale pixels at night) is used to track humans around the frame, but you can't search a video for a color in it.
The default setting for each camera added is to save clips whenever Any Object appears in the frame. That could be a leaf, a shadow, rain, a dog, a car, anything. Sighthound won't distinguish between them, it will just say Object Seen. If you change the setting to People, it will try to find only the objects that look very similar to the humans it was trained on. Does it get it right all the time? No. But it is successful enough that people don't get alerts from leaves, rain or dogs most of the time.
On Night Detection - if you have a decent IR camera, Sighthound will be able to see the scene almost as well as during the day. The tracking still works in gray scale, and the people detector works well.
On bandwidth we tend to see the video encoding and decoding slowing down the CPU before we see network issues. I'm sure if you add enough Hi Def cameras over Wifi and there is other stuff going on in the household, you'll clog your Wifi, however we have users with 25 cameras running on a big server with no network issues.
Stephen
-
Ronburgundy415 - Posts: 2
- Joined: Fri Oct 03, 2014 9:46 pm
Re: SightHound features list comparison?
To clarify, by direction I meant say something that moves across the screen vs. something that is moving around in one spot (squirrels fighting, human loitering, dog sleeping, etc). Can it distinguish that?
So it can remove false positives like wind/leaves, rain, squirrel, bugs, etc? There is a People setting that lets you choose to see only clips with people in it? What % accuracy do you think it is?
Lastly, how is it in crowded scenes? Say you have 5 cars, 5 people, 5 animals, and 5 other sizable objects. Some are moving and some are not. Can it detect all the items, and distinguish which is which?
So it can remove false positives like wind/leaves, rain, squirrel, bugs, etc? There is a People setting that lets you choose to see only clips with people in it? What % accuracy do you think it is?
Lastly, how is it in crowded scenes? Say you have 5 cars, 5 people, 5 animals, and 5 other sizable objects. Some are moving and some are not. Can it detect all the items, and distinguish which is which?
-
ryan - Posts: 1015
- Joined: Wed Aug 25, 2010 2:52 pm
- Location: Palo Alto, California
Re: SightHound features list comparison?
Hi Ron,
You can create a rule that responds to objects crossing a boundary in a given direction. Such a rule will fire each time an object crosses that boundary in the given direction. A detail of all rule options can be found in the following documentation: https://www.sighthound.com/support/rule-editor
Recognition accuracy, behavior in crowded scenes, and similar metrics are all difficult to say. We feel that our software beats everything out there, but in the infinite possibility of configurations cameras and scenes there will undoubtedly be some it does not perform as well in. The software includes a 14 day Pro trial, I'd highly recommend that you download and give it a shot to see if it meets your needs.
Best,
- ryan
You can create a rule that responds to objects crossing a boundary in a given direction. Such a rule will fire each time an object crosses that boundary in the given direction. A detail of all rule options can be found in the following documentation: https://www.sighthound.com/support/rule-editor
Recognition accuracy, behavior in crowded scenes, and similar metrics are all difficult to say. We feel that our software beats everything out there, but in the infinite possibility of configurations cameras and scenes there will undoubtedly be some it does not perform as well in. The software includes a 14 day Pro trial, I'd highly recommend that you download and give it a shot to see if it meets your needs.
Best,
- ryan
Learn more about Sighthound Video in our support pages - Reference Guide | All Articles
Are you a developer? Check out our cloud APIs - Demo | Docs
Are you a developer? Check out our cloud APIs - Demo | Docs