VCA Capabilities per Device - FW 6.30

VCA Capabilities per Device - FW 6.30
Bosch Security Systems
Technical Note
VCA Capabilities per Device FW 6.30
April 21, 2016
VCA Capabilities per Device
Different devices, whether cameras or encoders, offer a different amount of processing power for video content analysis
(VCA). This has effects on the supported feature set and detection distance. All Bosch IP cameras & encoders can be
clustered by their “Common Product Platform” (CPP), so the CPP version will be used to summarize the differences
between the versions. Further differences are introduced because new features are often added with FW updates,
because thermal camera videos have different visual properties than those of normal cameras , and because moving
cameras (AUTODOME/MIC) sometimes need different handling. In this tech note, all of these differences are collected.
Table of contents
(1) Overview of video content analysis algorithms
(2) Intelligent Video Analytics / Essential Video Analytics capable devices by common product platform (CPP)
(3) Resolution: MOTION+
(4) Resolution & minimal object size: Essential Video Analytics
(5) Resolution & minimal object size: Intelligent Video Analytics
(6) Frame rates: Intelligent Video Analytics / Essential Video Analytics / MOTION+
(7) Moving camera (AUTODOME / MIC) & VCA: Particularities
(8) Thermal camera (VOT) & VCA: Particularities
(9) VCA features: Differences by device
(10) Intelligent Video Analytics features by version
2 | Bosch Security Systems
(1) Overview of video content analysis algorithms
There are different types of video content analysis (VCA) algorithms available in different camera value levels. The camera
value level is always part of the camera name, e.g. the DINION IP starlight 8000 MP has a value level of 8000. Here the
overview of the VCA algorithms for FW 6.30:
Tamper detection
1000
4000
7000
-
-
-
3000
6000
9000
x
x
x
Description
•
Detect if camera is obscured or moved
•
Evaluate overall lighting level (too dark / bright)
•
Detect removed or placed objects in dedicated areas
(reference image check)
MOTION+
Essential Video Analytics
(CPP7 only)
Intelligent Video Analytics
x
x
x
x
x*
x
Intelligent Tracking
(MIC/AUTODOME/MP
cameras only)
x
•
Change detection to trigger recording
•
Intrusion detection for smaller, controlled environments
•
Counting / queuing applications
•
Mission critical perimeter protection
•
Outdoor environments with harsh weather conditions
•
Long detection ranges
•
Ship tracking
•
Exhibit protection
•
Counting / queuing applications
•
AUTODOME / MIC automatically zoom to & follow a selected
object
•
Megapixel cameras: “Virtual PTZ”, cutout stream follows
selected object in the camera field of view
* Intelligent Video Analytics includes the full Essential Video Analytics feature set
©2016 Bosch Security Systems
3 | Bosch Security Systems
(2) Intelligent Video Analytics / Essential Video Analytics capable devices by common product platform (CPP)
Below is a table of Intelligent Video Analytics (IVA) capable devices from Bosch with respect to their Common Product
Platform for CPP3, CPP-ENC, CPP4 and CPP6. For many of the devices listed there, IVA and non-IVA variants exist. From
CPP7 onward, Intelligent Video Analytics is part of all 7000-9000 value level cameras, whereas Essential Video Analytics is
part of all 4000-6000 value level cameras, therefore the devices are not listed explicitly there.
The CPP version of any device is part of the device type, which can be found on the device web page under system
overview, or in the configuration manager under general -> device access -> camera identification.
Note that though CPP3 is no longer supported and not included in FW 6.30, it is still listed in this tech note for
completeness.
CPP3
Fixed Cameras
PTZ Cameras
Encoder
DINION NBC-455-P IVA
AutoDome Easy II IP
VIP X1XF
DINION NBC-498-P IVA
AutoDome Jr 800 HD
DINION NBN-921-P IVA
HD Converence Dome
FLEXIDOME NDC-455-P IVA
AutoDome Jr 800 HD Fixed
FLEXIDOME NDC-498-P IVA
VG4 AutoDome
FLEXIDOME NDC-921-P IVA
AutoDome 800 HD
VOT-320V IVA
AutoDome 700 IP
NEI-30 IR Imager IVA
MIC VIP
CPP-ENC
CPP4
VIP-X16000-XFM4
DINION IP 7000 HD
AUTODOME IP 7000
DINION IP dynamic 7000 HD
AUTODOME IP 7000 HD
DINION IP starlight 7000 HD
AUTODOME IP starlight 7000 HD
FLEXIDOME IP 7000 RD
AUTODOME IP dynamic 7000 HD
FLEXIDOME IP dynamic 7000 RD
MIC IP starlight 7000 HD
FLEXIDOME IP starlight 7000 RD
MIP IP starlight 7000 XT
FLEXIDOME IP 7000 VR
MIC IP dynamic 7000 HD
FLEXIDOME IP dynamic 7000 VR
MIC IP dynamic 7000 XT
FLEXIDOME IP starlight 7000 VR
DINION IP imager 9000 HD
EXTEGRA IP starlight 9000 FX
EXTEGRA IP dynamic 9000 FX
DINION IP starlight 8000 MP
CPP6
DINION IP ultra 8000 MP
FLEXIDOME IP panoramic 7000 MP
©2016 Bosch Security Systems
4 | Bosch Security Systems
(3) Resolution: MOTION+
MOTION+ aggregates pixel information into processing blocks. These blocks are, in the internal resolution used for
processing, either 6x6 pixel large with 1 pixel overlap to the next blocks, or 8x8 pixel with no overlap to neighbouring
blocks, depending on whether the device was IVA capable until FW 6.10. From FW 6.10 onward, all devices now use the
8x8 pixel block raster. At the image border, there is an insensitive area with a width of approximately one block.
MOTION+ blocks
6x6 block raster, 1 pixel overlap
8x8 block raster w/o overlap
Internal resolution, block size and number of blocks used for MOTION+:
Platform
Aspect
Ratio
Internal Resolution
320
38
38
320
240
38
28
16:9
320
180
38
21
IVA capable
16:9
320
180
38
21
or
5:4
255
204
30
24
Not IVA capable
and < FW 6.10
CPP 3/4/ENC
CPP 3/4/ENC
1:1
320
4:3
Size
y
≥ FW 6.10
Y
Number of Blocks
x
CPP6/7
x
Block
8x8
4:3
240
180
28
21
PAL
176
144
20
16
NTSC
176
120
20
16
VGA
320
240
38
28
16:9
160
90
29
15
VGA
160
120
29
21
PAL
176
144
33
26
NTSC
176
120
33
21
6x6
(4) Resolution & minimal object size: Essential Video Analytics
Essential Video Analytics usually subsamples the video before processing it, resulting in the following internal processing
resolutions:
Platform
Aspect Ratio /
CPP7
©2016 Bosch Security Systems
IVA Resolution
Comment
Format
x
y
4:3
320
240
3:4
240
320
16:9
320
180
9:16
180
320
DINION / FLEXIDOME: 90° / 270° image rotation
DINION / FLEXIDOME: 90° / 270° image rotation
Any object with an area of less than 10 square pixels in this internal resolution will be discarded. A minimum object area of
20 square pixel is recommended for object detection.
(5) Intelligent Video Analytics resolution & minimal object size
Intelligent Video Analytics usually subsamples the video before processing it, resulting in the following internal processing
resolutions, which depend on camera platform, firmware version and configuration of Intelligent Video Analytics:
Aspect
Platform
CPP6/7 ≥FW 6.30 &
double distance active
CPP6/7 ≥FW 6.30 &
double detection inactive
CPP6/7
<FW 6.30
CPP3/4/ENC
IVA Resolution
Ratio /
Comment
Format
x
y
1:1
640
640
4:3
640
480
3:4
480
640
16:9
640
360
9:16
360
640
DINION / FLEXIDOME: 90° / 270° image rotation
1:1
320
320
FLEXIDOME IP panoramic 7000 MP 360°
4:3
320
240
3:4
240
320
16:9
320
180
9:16
180
320
DINION / FLEXIDOME: 90° / 270° image rotation
1:1
320
320
FLEXIDOME IP panoramic 7000 MP 360°
4:3
320
240
16:9
320
180
16:9
320
180
4:3
240
180
PAL
176
144
NTSC
176
120
VGA
320
240
FLEXIDOME IP panoramic 7000 MP 360°
DINION / FLEXIDOME: 90° / 270° image rotation
DINION / FLEXIDOME: 90° / 270° image rotation
VOT-320
Double detection distance is active on CPP6/7 for the following configurations & objects:

Tracking modes: 3D tracking, 3D people tracking

Noise suppression: OFF, MEDIUM

Object type: Moving, started, stopped
Any object with an area of less than 10 square pixels in the IVA resolution will be discarded. A minimum object area of 20
square pixel is recommended for object detection.
(6) Frame rates: Intelligent Video Analytics / Essential Video Analytics / MOTION+
Intelligent Video Analytics, Essential Video Analytics and MOTION+ are always processed on the first stream. Depending
on the maximal possible frame rate, or frames per second (fps) of this stream, they are processed with the following frame
rates:
Maximal
Intelligent Video Analytics /
fps
Essential Video Analytics
1rst steam
fps
25 / 50
12,5
5
30 / 60
15
5
8,33
8,33
5
MOTION+
fps
6 | Bosch Security Systems
If the actually used frame rate of the first stream is lower, then the metadata and alarms are still generated with the frame
rate specified here. If such a recording is viewed, then the displayed metadata will change faster than the video images
themselves.
There are only two cases where the frame rate drops below:

Exposure time is too large. This can be avoided by ensuring that minimal frame rate in automatic exposure mode is
not smaller than the specified Intelligent Video Analytics / Essential Video Analytics / MOTION+ frame rate.

Intelligent Video Analytics / Essential Video Analytics has to track too many objects simultaneously. This can happen
with 3D person tracking mode in crowds. At the moment, ~10 objects can be tracked in real time with Intelligent Video
Analytics on CPP4, ~20 objects with Intelligent Video Analytics on CPP6/7 and ~10 objects with Essential Video
Analytics. Beyond that, frame drops may occur because processing took longer than available for the frame.
Nevertheless, Intelligent Video Analytics / Essential Video Analytics will still continue to track the objects.
(7) Moving camera (AUTODOME / MIC) & VCA: Particularities
Many VCA algorithms are designed for non-moving cameras only, and will not work while the camera is moved. Therefore
these algorithms are only available on predefined dome scenes. They will start the moment the scene is reached and be
stopped on departure.
(7.1) Profiles vs Scenes
On non-moving cameras, up to two VCA profiles are supported, which can be scheduled or switched via an internal or
external trigger event. Moving cameras, however, are typically used different than non-moving cameras. They are typically
configured to move to different scenes after a certain time span, in order to monitor more area than with a single camera.
This would collide with any attempt to schedule the VCA on a single scene. Thus, while up to 16 scenes are supported with
VCA, only a single VCA profile is available on each scene, and scheduling is not available.
(7.2) Reference Image
Due to limited on-board storage and the fact that not only two but 16 reference images need to be stored on moving
cameras, the VCA algorithms relying on a reference image, that is reference image check for tamper detection and crowd
density estimation, have long not been supported and are only available from FW 6.20 onward on CPP4 an d newer.
(7.3) Intelligent Tracking
Moving cameras have a special VCA algorithm called Intelligent Tracking, which is only available on them: When triggered
by an Intelligent Video Analytics or the manual selection of an Intelligent Video Analytics object by a user, the Intelligent
Tracking steers the moving camera all by itself to zoom in on the object for most details and to follow this object. When the
object stops, the Intelligent Tracking will monitor this region for a user specified time for further movement to be picked up
and tracked. Afterwards, the moving camera will return to the scene it came from.
(8) Thermal camera (VOT) & VCA: Particularities
Thermal cameras capture light not in the visible, colourful spectrum, but as heat emissions. Therefore no colour information
is available. Furthermore, areas and object with same temperature will have same thermal values, and thus distinct scene
properties in the visible spectrum like fences, walls and bags may not be visible in the thermal image at all. Thus tamper
detection and all reference image check based algorithms, that is reference image check and crowd density estimation, are
not supported either, nor is the detection of idle / removed objects.
©2016 Bosch Security Systems
7 | Bosch Security Systems
Alarm Rules
Object Filter
Tracking Modes
✓
✓

✓

✓
✓
Tamper detection
✓

✓
✓
✓
✓
✓
Reference image check
✓


✓
✓
✓
✓
MOTION+
✓
✓
✓
✓
✓
✓
✓
Essential Video Analytics
CPP7
Intelligent Video Analytics
CPP6/7
Intelligent Video Analytics
CPP4 PTZ
Intelligent Video Analytics
CPP4
Intelligent Video Analytics
CPP3 VOT-320
VCA scheduling
Feature list
based on
CPP3: FW 5.50
CPP4/6/7: FW 6.30
Intelligent Video Analytics
CPP3 PTZ
Intelligent Video Analytics
CPP3 / ENC
(9) VCA features: Differences by device
Intelligent Tracking


✓

✓


Any object
✓
✓
✓
✓
✓
✓
✓
Object in field
✓
✓
✓
✓
✓
✓
✓
Line crossing
✓
✓
✓
✓
✓
✓
✓
Enter / leave field
✓
✓
✓
✓
✓
✓
✓
Loitering
✓
✓
✓
✓
✓
✓
✓
Follow route
✓
✓
✓
✓
✓
✓
✓
Idle / removed object
✓

✓
✓
✓
✓
✓
Counting
✓
✓
✓
✓
✓
✓
✓
Occupancy
✓
✓
✓
✓
✓
✓
✓
Crowd density estimation
✓


✓
✓
✓
✓
Condition change
✓
✓
✓
✓
✓
✓
✓
Similarity search
✓
✓
✓
✓
✓
✓
✓
Flow / counter flow
✓
✓
✓
✓
✓
✓

Duration
✓
✓
✓
✓
✓
✓
✓
Size
✓
✓
✓
✓
✓
✓
✓
Aspect ratio
✓
✓
✓
✓
✓
✓
✓
Speed
✓
✓
✓
✓
✓
✓
✓
Direction
✓
✓
✓
✓
✓
✓
✓
Color
✓

✓
✓
✓
✓
✓
Object class



✓
✓
✓
✓
Geolocation



✓
✓
✓
✓
Standard (2D) tracking
✓
✓
✓
✓
✓
✓
✓
3D tracking
✓
✓
✓
✓
✓
✓
✓
3D people tracking
✓
✓
✓
✓
✓
✓
✓
Ship tracking





✓

Museum mode
✓
✓
✓
✓
✓
✓
✓
Masking
✓
✓
✓
✓
✓
✓
✓
Face detection



✓
✓
✓

Scene-specific Object Detection (via CBS*)





✓

*Bosch Cloud-based Services (CBS)
©2016 Bosch Security Systems
8 | Bosch Security Systems
(10) Intelligent Video Analytics features by version
Here is an overview of which features were introduced by which Intelligent Video Analytics (IVA) version, starting with IVA
4.0 as the oldest IVA version supported by the devices listed here. The supported common product platforms of the FW /
IVA release are added as well.
IVA Version New Features
CPP-ENC
CPP3
CPP4
x
x
x
x
x
x
x
x
CPP6
CPP7
IVA FLOW
4.0
Head detection
Improved calibration tool
4.5
Crowd detection
Improved object detection & tracking
Counter
5.5
People counter
Assisted self-calibration wizard
Improved object detection & tracking
5.6
Best face detection (CPP4 only)
x
Added CPP6 with full CPP4 feature set
6.0
Object classification (replaces head detection)
x
Improved configuration
Object classification (replaces head detection)
Improved configuration
6.1
Ship tracking (CPP6 only)
x
x
x
x
x
x
Geolocation
Improved object detection & tracking (CPP6 only)
Improved IVA Flow (CPP6 only)
Improved robustness to shaking camera (CPP6 only)
6.2
Scene-specific Object Detection (CPP6 only, via CBS)
Reference image check & Crowd density estimation for
moving cameras (CPP4 only)
Added CPP7 with full CPP6 feature set
Improved calibration: Focal length from lens motor
position (CPP4-7), angles from gyrosensor (CPP7 only)
6.3
or PTZ position (CPP4-7)
Polylines for line crossing
Alarm on foot points
Alarm on number objects in field (now via GUI)
Readded museum mode (CPP6/7)
Bosch Sicherheitssysteme GmbH
Robert-Bosch-Ring 5
85630 Grasbrunn
Germany
www.boschsecurity.com
©2016 Bosch Security Systems
x
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement