Given IP cameras use compressed video where compression artefacts might lead to false positives, I wonder how VMD accuracy would compare between equivalent IP and HDCVI cameras from Dahua? HDCVI sends uncompressed video to the DVR which should be better for VMD analysis except for any analog noise. I wonder if there would be any difference and, if so, whether it would be a little or a lot?
VMD Shootout II - Hikvision, Dahua, Samsung
*** ******** ****** **************** ********, ****, ***** *** ****. Members ******* ** *** ***** *** more ************* ** ** ******. **** test ******** * *** ************* - Dahua,************ *******, ******** ** *** ******** camera **** *******:
** ******** *** **** ******* ******** as *** ***** **** ******, ******** **** a ****** ****** ********** ****:
****** ** ** * **** ******* open ****** ****:
*** *******, ** ******* **** **** moving ********** *** ********** ** * nearby ****:
*** ***** ****** ** ****** *** following *** *********:
- **** ****** **** ******** ***** ***** positives?
- ***** *** *** **** ** ********* these ******?
- *** *** ********* ***** **** **** manufacturer ** ************?
- *** ******** ** ********* *** ****** detection ******* ********** ***** ****** *** vice *****?
Key ********
***** ****** **** **** ***** ***** were ********* ***** ************* * ******* 60° ********** ***** ** **** **** top ** *** **** *** ******* from *****,*********, *** *******. ** ********* **** preliminary ******* **** ******** ****** *** found*********** *** ***** ** **** ******, but ** ****** *** ** ******* that *** ******* ** *** ************'* line ******* ********* ** *** ****** models.
**** ******* ** *********** *** ******** (********, ****, *****, ****) ******* ***** ****** ********* *.*, ***** ***** ************* ********* ** *******:
- *****'* *********** *** ************* *** **** as **** (******** ***), *****, *** Sony ****** *** *** ** **** range *** ***** ********** *********. ** night, ** ************ **** ******** *** and ****, ********* ***** ***** ** *** ***** NBN-733V, **** **** *** ***** ***********.
- ********************* **** (******** ***), *****, *** Sony ****** *** ***, ** **' range **** *** ***** *********. ********* range ** ***** *** ****** **** all **** ************* ** *** ******** test, ** **', ******* ******* ***** positive ******.
- *******'* ***** (~***' ********) *** ******* than *** *******, ********* ******** (~***'), the ******** *** *********, **** *** false ***********. ** *****, ********* ***** and ******** *** **** ****** (~**' vs. **' *** ********).
- ******** ** **** *** *.*, *********** varied ********* ** *** *****. *** three ************* ** **** **** ************ Axis *** *.* ** *** *****, where ** ****** ******** ********** ** even ***** *****. ****** *** ***, Samsung ************ *** *.* ** ~**-**',************ ************* ***** ** ***** (~**-**'), while ***** *** ~**' *******. **** VMD *.*'* ***** ******** ********* **** was ****** *******, *******, ****** **** any ****** ** **** ****.
***** *** ******** ******* ********** ****** performance *** ******:
- ****** *****'* ***** *** ** *** with **** ***** *******, *** ***** activations **** *** ** *****, ** found ***** ****** ********* ** ** the ***** ********, **** ******. ** failed ** ******** ************ ** ***** despite *** ***** *** ******** ********* unchanged.
- *** ******* ***-**** *** *** **** detection ***** ** **** ****, **** indoors *** ***, ** **** ***** and ****. *******, ** *** **** prone ** ***** *********** ********, **** shadows *** ******* *** ******, ** flashing **** ** *** **** ****** test *****.
- *********'* *********** *** **** **** ** ******* in ******** ******, ** *** ***** levels ******. ********, ********* ***** *** shorter **** *******, ****** ****** **** Dahua. ***** *********** **** ********, **** the ********* ** ******* ****** ** the *****, ***** ************ ********* ****** detection.
Optimization ***************
***** ** *** *****, ***** *** our *************** ** *** ** **** optimize **** ******:
- ********* ******* ****** *** *********** * (*******) or *. ***** ****, ***** ****** are ********** ******, ***** ***** *********** increase *********** ***** *.
- ** ********* ******* *****'* ******* *********** ** 55, **** * ********* ** *-*. Increasing *********** ***** **** ******* ** numerous ***** ************. ** **** ******, ***** ******** zones ** ***** *** ****, ******* of * ****** ***** **** ********* better, *** ** *** *** **** consistent ******** ** ** **** **** is **********.
- *** *******, ** ***** **** ******* the *********** ** *** ******* "******" setting *** *** **** ***** ** true ********** ** ***** ******, *** lowering ** ******* *********** **** ****** directly ****** ********* ***** ** ******** the ********** ** ***** ******.
*************
*** ********* ****** ****** ************* ** each ************'* ***.
***** *******
********* **-*********
******* ***-****
Outdoor *******/*********
******* ********* *****
** ~***', *** ******* ***-**** ** the **** ****** ** ****** *** subject, ****** ** *** ******* ********* range ** *** *** ******* ******.
** ~**', *** ********* **-********* ****** ** ****** *** ******* consistently.
*******, ** ~**', *** ***** ******* is **** ** ****** *** *******.
********* ********* *****
*** ** *** ******* ****** *** a ********* ** ********* *****, **** the ***** ***** **** ~**' ** *** ******* to ~**' ** *****, ********* ***** **** ~**' ** ~**', *** Samsung ***** ****** *** ******** ********* range ** *****, *** ******** **** ~145' ** *** *** ** ~**' during *****.
Indoor ********** ****
** *** ********** **** ** **** light, *** ******* **** **** ** reliably ****** * ***** ******* ** all ****** ** ** **'.
** *** ***** (~***), *******'* *** was *** **** ****** ****** ** reliably ****** *** ******* ** *** ranges ** ** **' ** *** conference **** *****. *** ********* ******* *** ** **% ******** ****** with *** ******* ** ~**', *** 100% ********* **** *** ******* *** ~18' ** ******. ***** ******* ********* poorly ** **** *****, ******** **% detection **** *** ******* ~**' ****, and ******** ***% ********* ******** **** our ******* ~**' **** *** ******.
*** ********* ******* *** * **% *****-***** **** with * ****** ****** ** *** scene, ***** *** ******* ***-**** *** a **% *****-***** ****. **** ***** cameras *** *** *******.
**** ** ***** * ********** ******* the ****, *** ***** ******* *** a **% *****-***** ****, ***** *** ********* *** ******* ******* **** **********.
******* *** **** ******* *** ***** ** the ***** *** ******* * ***** alarm *************** **% *** ******* **%. ***** ******** unaffected.
Open ******
****** *** ******* ** ** **** office *****, ***** **' ****, *** our ******* ~** **** *** *******, full ***** *********** ** ********* *** ******* *** *******, **** ******* detecting *** ******* ** **** ** he **** ****** *** ******, *** the ********* ************ ** *** **** ******* ~**% of *** ****. **** *** ***** box *** ******** ******* ***** *** reliably ****** *** ******* **** **** distance.
*******, **** *** ******* ****** ** move **** *** *****, *** ******* are **** ** ****** *** *******, with ********* ******** *** ****** ** walked ** *** *******.
** *** **** ***** ** *** light, *** ********* *** ******* ******* ****** ***** ******* detection ***** ** * **** ***** scene, *** ***** ** ****** ** detect *** *******.
*** ******* ***-**** *** * **** frequency ** ********* ****** ******** ** a ******** ***. ***** **** ******* 20 ****** ***** ****. *** ********* *** ***** ******* *** *** **** false ****** **** *** ******** ***.
Test **********
*** ********* ******* *** ******** ******** were **** ** *** ****:
- ***** *******: *.***.******* **.*.*
- ********* **-*********: **.*.*
- ******* ***-****: *.*********
Motion detection is always done on compressed video. Also, compression artifacts, assuming a 'regular' compressed video (not one that has been super compressed for a tiny bit rate), should not be a factor in detection accuracy.
I think this report is due for an update, especially since you are using it in the camera course. Moore's law alone should suggest an update is due to keep the report relivant. Also could you test VMD on panoramic cameras and some American made cameras?
#1, thanks, we will schedule some more VMD testing but, even from what vendors are claiming, Moore's Law has not a big impact on VMD.
There is the deep learning NVR coming from Hikvision (soon) and we definitely will test that. And Axis has VMD4 which they claim has improvements which we will test. However, overall there is not many VMD advance claims.