Lilin YOLO-Based Analytics Tested
While YOLO is one of the most popular object detection algorithms, few use it in commercial video surveillance. Now, Taiwan manufacturer Lilin is equipping its Z7 line with YOLOv3, but how well does it work?
We bought and tested the Lilin Z7R6482X3-P, and an AI-EDGEFLOW license examining:
- How well does it ignore common sources of indoor and outdoor false alerts?
- Does heavy rain cause false alerts?
- Can people be detected in different postures?
- Do false alerts occur on random objects in the scene?
- How simple or difficult is the setup?
- How well is integration with 3rd party VMS?
- How accurate is vehicle detection?
- Are parked vehicles an issue?
Executive *******
*** **** ******* ** *****'* ********* performance ***** ** ***-****** *************. ** excelled ** **** * *** ********** while *************** ** ***** ******** *****.
** *** ******** ****, ** ******** people ** ********* ********, ******** ***** alerts ** *******, ****, *** ****** objects, *** ********** ********** ********.
*******, ** ********* ******* ******* ** static *******, ******** * ****** *** for ****** *********, *** **** ****** tracking, *** ************ *** ***********, *** had ******** ******* *********.
***** ******* **** ** ***** ** our****** ********* ******** **** - ** Manufacturers, ** ************** **** *******.
Pricing ************* **** **** ***********
*** *******-***************** *** ** ********* *** $*** per ******* *** ******** ****** ** ****** *********** *** **** ** ********* *** ~$600 - $*** ****** *** * Z7 ****** ******* ***. **** ** significantly **** ********* **** *********** **** Hikvision *** *****, ***** **** ******* analytics ******* **** ** *** ******* an ********** ******* *** ~$***, *** costs **** **** ****' ~$****-****** ****** **** **** *********.
*****
******* ***** *** ***** ** ******, only ********* *** ****** ** ******** to ** ***** *** ***** ******* will ** ********. ****** ******* ************ will ** ******** ** ****** ***** events, **** ** ****-********* *** ********** due ** *** ****** ** ************* tools ******, **** ** ********** ***** adjustments, **** *****, *** ******* ***** alerts.
** ****** **** ** *** ***** below:
Constant ****** ** ****** ********
** *** *******, ***** ******** **** constant ****** ** ****** ******** ** the *****, ***** ****** *** ***** list **** *** **** ***** *** flooded *** ****** **** *** **** detections ** ******* **** *********.
***** *********** ******** ** ** * false ***** ** *** *** **, but **** *** *** *** *** issue. ******* ********** *** ** ** turned *** ********** ** *** ****** was ******* ** ********** ********.
Random ******* ******** ** ******
***** ******** *** ****** ********* ****** objects ** ******, **** ** **** Non-swaying ****, * ***** ** *** distance, *** * ******* ******* ****.
***** ********* ***** *** ****** ** flag * ***** ***** ****** **** did *** *** *** ** ****.
Low-Resolution ********** **** ********* *******
****** ****** *** ***** *** ******** low-resolution ****** ** ****** **** ********. This *** ** ***** ****** ** search ******** ******** *** ****** ** identify ************ **** ****** ** *****.
***** ******** ** **** **** ** an *****, *** **** **** ** fixing **** *** ****** *** ********** a ****** **********.
Accurate ******* ***************
******* *************** **** ********, ********* ****, Trucks, ********, *** ***** ** *** scene.
*******, ***** **** **** ***** ***************, as ****** ****** **** ********** ** cars ***************.
Poor ******** ** ******
**** * ****** *** ***** ** the ****** ** * **** ***, the ******** *** ***** ********** *** smoothly ***** *** ****** ** *** scene, ****** ******* **** ** * low *** (~**), *** ******** *** would ************ **** *** **** ** the **** ******, ******** ** * new ******.
People ********** ******** *******, ********, ******* ******
***** ********** ******** ****** *******, ********, and ******* ***** ******** ***** *******.
****** ********* ********** ** *******, **** as **** ****** **** **** ** their **** **********, **** ******** *** accurately ********** ** ******.
No ***** ****** ** ****** *******
***** *** *** **** *** ***** alerts ** ****** ****** ***** ******, such ** ***** ***** *******.
************, ***** *** *** **** ****** with ****** ******* ** *** ***** of ******.
No ***** ****** ** *******
***********, ***** ******* ******* ******* ******* the ***** *** *** *** ***** a ******** *** ** *** ******.
No ***** ****** ** ***** ****
***** **** *** ******** ******** **** the **** **** *** ** ***** for *****. ***** **** ** ***** classifications ****** ***** ********* ****** ~* weeks ** *******.
People ******** ** ~**-** ***
***** ******** ******** ****** ** ~**-** PPF, ****** **** *********** **** ** Avigilon ** *****, ***** **** **** ~8-10 ***.
No *** *********** - ***** *** ******** *** ***** *******
***** **** *** ******** ********* *** analytics **** *** ***** ***. *******, the **** **** *** ** **** events ** *** *****'* *** *** to **** ****** *** ****** ********.
******** ****
*** ********* ******** **** **** ****** testing.
- ***** *********-*: **.*.***.****
- ***** ********: *.*.*.**-******