Concept Pro DEEP LEARNING PLUS User manual

QUICK START GUIDE
DL PLUS-4/DL PLUS-8/DL PLUS-16
Technical Documents
DEEP LEARNINGDEEP LEARNING PLUSPLUS

DEEP LEARNING PLUS QUICK START GUIDE2

3DEEP LEARNING PLUS QUICK START GUIDE
WHAT’S IN THE BOX?
Packaging Contents
Technical Documents
DL PLUS Unit x1 Quick Start Guide x1
DC12V Power Supply Adapter x1 Cable Clamp x1
Screw x1
INSTALLATION
REAR PANEL
DL PLUS-4
DL PLUS-8 & DL PLUS-16
No. Item Description
1 USB Universal Serial Bus (USB) ports for additional devices
(e.g., USB storage device for firmware upgrade).
2 HD MONITOR HDMI port for connection to monitor to view live video. Note: DL PLUS cannot
be configured locally.
3 ETHERNET 1 RJ-45 port for network connection. Allows DL PLUS device configuration,
interoperation with VMS, video recorder, and IP camera equipment.
4 ETHERNET 2 RJ-45 port for connecting IP cameras and other equipment in a separate
network environment.
5 ALARM IN Alarm input signal terminal for connection with alarm sensor equipment.
6 RELAY Alarm output relay terminal for connection with external alarm equipment.
7 RS485 RS-485 communication terminal for connection with external equipment
(e.g., access control equipment)
8 DC12V DC12V power input for included power supply adapter.

DEEP LEARNING PLUS QUICK START GUIDE4
BASIC CONNECTION OVERVIEW
DL PLUS DECODING PERFORMANCE
The DL PLUS product supports the following decoding performance from video sources:
Max. Frame Rate @ Video Source Resolution
Model 2-megapixel 4/5-megapixel 8-megapixel
DL PLUS-4 4CH@100FPS 4CH@48FPS 4CH@24FPS
DL PLUS-8 8CH@200FPS 8CH@96FPS 8CH@48FPS
DL PLUS-16 16CH@400FPS 16CH@192FPS 16CH@96FPS
It is important that all channels used as video sources for DL PLUS meet this decoding specification
in order to prevent issues with product stability. If necessary, reduce resolution and/or frame rate for
video sources from their respective stream/record setup menus to ensure total decoding performance is
maintained.
For example, the following mixed load-out would be acceptable for DL PLUS-4:
Channel Resolution Frame Rate
Channel 1 1920x1080 25FPS
Channel 2 2592x1520 12FPS
Channel 3 2592x1520 12FPS
Channel 4 3840x2160 6FPS

5DEEP LEARNING PLUS QUICK START GUIDE
DL PLUS DETECTION PERFORMANCE
The DL PLUS products can support up to 80 total deep learning rules applied to up to 16 video sources.
This product is also currently capable of detecting humans, cars, and bicycles at higher speeds and
distances than traditional built-in video analysis.
Actual detection range of objects depends upon the rule type, target object, camera field of view, and
camera installation height/angle. Some general guidelines for minimum object size relative to total image
size are below:
Minimum Object Size for Intrusion & Occupancy Detection
Object Type Person Vehicle
Algorithm Width Height Width Height
Human/Vehicle – Far 1.00% 3.00% 2.00% 1.50%
Human/Vehicle – Normal 1.25% 4.50% 3.00% 2.25%
Minimum Object Size for Loitering, Enter/Exit & Line Crossing Detection
Object Type Person Vehicle
Algorithm Width Height Width Height
Human/Vehicle – Far 1.25% 5.00% 3.00% 2.50%
Human/Vehicle – Normal 2.00% 6.00% 7.00% 6.00%
For cameras installed at heights above 3m, tilt angle plays an increasingly crucial role in determining
detection accuracy. Some general guidelines for minimum object size relative to total image size are
below:
Minimum Object Size for Installations >3M High (Normal Algorithm)
Camera Tilt Angle Trigger Rule Person Vehicle
Width Height Width Height
Within -30°
Intrusion, Occupancy 1.25% 4.50% 5.00% 2.50%
Loitering, Line Crossing,
Enter/Exit, Counting 2.00% 6.00% 7.00% 6.00%
-30 to -45°
Intrusion, Occupancy 1.70% 5.00% 8.00% 3.00%
Loitering, Line Crossing,
Enter/Exit, Counting 2.50% 7.00% 10.00% 4.00%
-45 to -60°
Intrusion, Occupancy 2.50% 6.00% 13.00% 5.00%
Loitering, Line Crossing,
Enter/Exit, Counting 3.00% 8.00% 20.00% 7.00%
-60 to -75°
Intrusion, Occupancy 3.50% 5.50% 30.00% 9.00%
Loitering, Line Crossing,
Enter/Exit, Counting 4.00% 7.00% 33.00% 10.00%
PLEASE REFER TO THE HOW TO OPTIMISE CAMERA INSTALLATION FOR ACCURATE
DETECTION SECTION AT THE END OF THIS QUICK START GUIDE FOR MORE GUIDANCE ON
SETTING UP A CAMERA FOR >95% DETECTION ACCURACY.

DEEP LEARNING PLUS QUICK START GUIDE6
DL PLUS OPERATION MODES
The DL PLUS product allows for up to sixteen channels and up to 80 individual rules of Deep Learning
video analysis to be performed with standard RTSP video streams from IP cameras, network video
recorders, and/or digital video recorders. The DL PLUS product can be used in one of the following three
different operation modes depending upon the CCTV system and customer requirements:
COMPANION MODE
The most convenient method to utilise the DL PLUS product in a Concept Pro system is Companion
Mode. Simply connect the DL PLUS unit to the same local network as a supported Concept Pro network/
digital video recorder as indicated in the “Basic Connection Overview” section above. Then, add the DL
PLUS unit from the DVR/NVR user interface and configure Deep Learning algorithm-based rules with
specific alarm triggers for each channel directly from the recorder.
UNIVERSAL MODE
Alternatively, DL PLUS can be used in Universal Mode. In this mode, DL PLUS can be configured directly
from its plug-in free web interface to receive RTSP video streams from multiple sources (multiple DVR,
NVR, and/or IP camera devices). Universal Mode allows for these video sources to be configured with
more advanced setup and algorithm options as well as more diverse event action trigger options,
including the option to transmit event metadata back to supported Concept Pro video recorder products.
EVENT SERVER MODE
Like Universal Mode, Event Server Mode allows for the DL PLUS product to be added to existing CCTV
systems and integrate Deep Learning event metadata with third party software systems. Any device with
an RTSP stream can be connected as a video source, including current and legacy Concept Pro DVR/
NVRs, Aspect products, and third party DVR, NVR, and IP camera products. Rather than sending event
metadata back to a DVR/NVR, the DL PLUS provides deep learning video analysis to an existing CCTV
installation and can be configured to trigger specific event actions on top of the current system. These
can be custom e-mail event notifications, alarm relay output, RS-485 output, or specific network outputs
to integrated VMS products and HTTP/TCP commands.

7DEEP LEARNING PLUS QUICK START GUIDE
COMPANION MODE SETUP
CONNECTING DL PLUS WITH SUPPORTED CONCEPT PRO RECORDER
Please follow the steps below to connect a DL PLUS product with a supported Concept Pro IP/AHD recorder:
1. Install the DL PLUS product on the same network as the desired, compatible Concept Pro
video recorder (e.g., VUHDIP Series NVR).
Note: A firmware upgrade may be required for the video recorder to work with DL PLUS.
2. From the Concept Pro Video Recorder user interface, navigate to <MENU> - <SYSTEM
SETUP> - <CAMERA> - <ANALYTICS> menu.
3. Under the <ACTIVATION> tab, select the <DL PLUS SCAN> button to scan the network
for any connected DL PLUS products. The DL PLUS device MAC address should appear
under the DL PLUS column. Select the DL PLUS button for each channel that will be
configured to use DL PLUS algorithms. Click <APPLY>.

DEEP LEARNING PLUS QUICK START GUIDE8
CREATING DEEP LEARNING RULES
1. After completing the steps above, navigate to the <PROPERTY> tab and select a desired
video channel from the <CHANNEL> drop-down menu .
2. Refer to the top section to the right. Set <ACTIVATION> to <ON>, select a DL algorithm
(e.g., Human/Vehicle Detector), and select <EDIT> under <RULE ENGINE> to add, edit, or
delete Deep Learning rules for the selected channel. The following rule types are supported
by DL PLUS and Concept Pro recorders:
Rule Type Description
Line Crossing Detection Triggers when a target object is detected crossing a custom-drawn line;
can detect forward, reverse, or both directions.
Intrusion Detection
Triggers continually when a target object is detected entering and
remaining inside a custom-drawn area. Note: objects remaining inside the
area will continue to trigger Intrusion Detection alerts until they exit the
area.
Enter Area Detection Triggers once when a target object enters a custom-drawn area.
Exit Area Detection Triggers once when a target object exits a custom-drawn area.
Loitering Detection Triggers when a target object remains inside a custom-drawn area for a
configured time period.
Stopped Detection Triggers when a target object enters and stops inside a custom-drawn
area for a configured time period.
Object Counter
Triggers when a target object is detected a specified number of times; can
be configured to reset count and can be linked to up to two other rules
(increase count rule and/or decrease count rule).
a. DL PLUS supports human, car, and bike target object identification/detection.
b. Configure a CONFIDENCE THRESHOLD when creating a DL rule to limit event
detection to cases where the DL PLUS product is X% confident in its target object
identification.

9DEEP LEARNING PLUS QUICK START GUIDE
3. Select <PREVIOUS> and then <CALIBRATION> to perform 3D Calibration for the selected
channel. Although not required for operation, performing 3D Calibration can result in more
accurate event detection and allow for enhanced filtering of target objects. Calibration can
be performed by creating at least five icons of a standard height and positioning throughout
the field of view.
4. Select <PREVIOUS> and then <ENGINE/OSD OPTION> to configure on-screen display
options for the Deep Learning metadata.
a. Configure IGNORE STATIC OBJECT for the selected channel to prevent false or
continual alarms on stationary objects in the configured rule area. Setting this to HIGH
will result in the system being less sensitive to object detection (available for Intrusion and
Occupancy Detection rule types).
5. When all rule configuration is complete, click <APPLY> and repeat for any other desired
channels.

DEEP LEARNING PLUS QUICK START GUIDE10
CONFIGURING ALARM/DEEP LEARNING EVENT RECORDING
Using DL PLUS to detect specific target objects based on custom deep learning-based detection rules
can result in reduced storage requirements for a CCTV system by only using the highest recording quality
(resolution, frame rate, quality, etc.) for Alarm/Deep Learning events and using lower quality recording
settings for other modes. To take advantage of this, please follow the steps below:
1. From the Concept Pro Video Recorder user interface, navigate to <MENU> - <RECORD
SETUP> - <DL/ALARM RECORDING> menu.
2. Click on the timeline bar on the top and drag for a desired time range.
3. Configure resolution, FPS, quality, and audio for any desired channels.
4. Navigate to <SCHEDULE> tab and click and drag a desired activation schedule for this
recording mode.
5. When all recording configuration is complete, click <APPLY>.