Video Analytics Person / Face / Vehicle Guide
Person, face, and vehicle detection are the most commonly offered video analytics, but understanding how they work and what challenges can break them is not easy.
In this guide we examine:
- How to tell how well an analytics provider's person, face, and vehicle detection work.
- The different pixel density requirements for person, face, and vehicle detection.
- Person Detection performance and most common accuracy problems
- Face Detection: performance and most common accuracy problems
- Face detection vs facial recognition
- Vehicle Detection: performance and most common accuracy problems
This is part of our new Video Analytics Course starting in March.
Different ********** ** ******, ****, *** ******* *********
********* ******, *****, *** ******** *** be ******** ********* ** ******* ******** and **** ********. ******* ********-***** ********* (HOG, ****) ** **** ****** ** IP *******, ******* ** ** ****** to *******, ******** *** ****-**** ******, and ******** **** ********** **** **** learning *******. *******, ******* ******** ** often ************* **** ******** **** **** learning (******* ****** **** ********* *********** **-***** ****** ********* ******), ************ **** *** *********** ****** of **** *** ******** ****** **** video ************ *****.
Performance ****** *** ********* *********
***** ********* *** ** ************* ****** or ***** **** *** ** **** to ********* *** **** ******** ****. While ************* ***** ****** ******** (*.*., up ** **%, **%, ***.), *** testing ***** **** ********* ** *** match ****** ******* ***********. *** **** way ** **** *** ******** ******** is ** ******* ** ******** ********** (e.g., **** **** ****).
*** ********* ****** **** *** ******* insight ** **** *** ** ****, but **** ************* ** *** ********* or ***** **** ***********. ********, ** can ** ********* ** *********,***** ********* *********** / ******** *** camera, *** **** ** *** **** the *** **** ***** ** ************ since ********* ********** ** ******** ***** varies.
** **** ******* *****, ****** ************* routinely ****** ******* ** "*****" ** having "**" *********, **** *** ****-***** testing ** **** *********** ******** ********. For ******* *******, ******* ****** ********* ********.
Pixel ******* (**********) ******** - ************ **** ******
********* ***** ******* ~*** ******** ***** of **** **** ********* ******. ******** can ** ******** ** **** ****** for ******.
*** ********, **** * ***** ******, faces *** ** ******** ** ** ~20' / ** ****, ******* ****** can ** ******** ** ** ***' / *** **** *** ******** ** to ***' / **** **** (******** no ************, **** ********, ***.)
Person *********
****** ********* ** *** ** *** most ****** *** ******** ***** *********, offering ****-**** ******** *** ******* **************.
********, ********* **** ** ****** ** a ****** (*.*. *** ** ******, vehicle, ** ****** *****) ******* **** accurate ***** ******** ********* **** ********* or ********/********* *********:
*******, ***** *** ************* *** ********** challenges ** ********* ******.
Person ********* **********
***** **** ****** ********* ********* *** accurate ** **** ***** **** ***** details ** * *******, ***** *** factors **** ***** ******. *** ** uncontrolled ******** *** ******* **********, **** of ***** ******* *** ************ ******* to ******* *******.
******* ** *** ** *** **** common ********** *** *********, ******** ********* snow *** **** ** ******:
***** ******* ** ******** ** ****** lights (*.*. ********** ** * ****** vehicle) *** ******* ****** ********* *** person ********* *********, ***** ***** ***** detections:
******* ** ******* ********** ** **** cameras, **** ****** ********* ********* *** miss ********* * ****** ******* ******* the ***** ** ******** ** * low ***** **** (* - * fps) *** *** ****** **** ** "see" *** ****** ****** ** * person:
**** ****** ********* ********* *** ********** or ***** ** ****** ****** **** are ******* (********, *******, ***.), ***** will **** ****** ** **** *** crawling:
**** ******* ** ****** ********* ***********/********, if *** ******** **** *** "***" the ****** ******, *** ******* *** not ** ********, ** ********* ** lost **** **** *** ********* ****** or ********:
* **** ****** ******* **** ***-********** analytics, **** ******* **** *** ******* and ****** **** ** * ******:
Face *********
***** **** ********* ** ********** ******, finding *** *********** ** ** ****** in * ***** ** * **** is **** *********** **** ******* * person ** ******* ******* * **** is **** *******. ********, **** ***** surveillance ******* (*.*., ** *******, ****) have ******* ********** *****, ***** ***** more ********* *** ***** ***** ******** analytics.
*********** ****** ** * *********** *******:
- ***** ** *****: ***** ** ** 'easy' ** ****** * **** ******* directly ** *** ******, *********** *** vary ************* ********* ** *** * person ***** ***** **** (****, ****, right, ***.)
- ******** ** ***** - ***** ** is '****' ** ****** * **** looking ******** ** *** ****** ** a ****-*** *****, *********** **** **** significantly ********* ** *** ******** ********** of *** ***** (*******, ********, *****, etc.).
******, ****-*** ***** *** ******* ** detect:
**** **** ****** *********, **** ********* performance ** **** ************* ******** ** environmental *** ********** **********.
Face ********* **********
***** **** **** ********* ********* *** accurate **** ****** ***** *** **** lighting, ******* ** ***** **** ************ cameras *** *********, ***** *** ********** problems **** ***** ******; ********* ***** and ********.
**** ******* *** ********* ** ********, and ******* *****/** ******* ** *****, resulting ** ******* **** ****** **** mostly *** *** ** *** **** visible:
** **** ******** * *********** ** the **** ** *** ****, ** a **** *****, ******* ** ***** cameras *** ********* *** ****** ****** look ******** ** *******:
******* *********** ******* *** **** *********, because ** *** **** ****** ********, is *** ** ****** ********. ***** a ****** *** ** ******* *******, low ***** ***** *** ******* *** face **** ****, *****, *** *********:
***** *** **** ********* *** *** least *********** *** **** *********, *******, can ***** ***-**** ******* ** **** faces **** **** *** ****** *** full ********:
*** **** ****'***** ********* ********.
Face ********* ** *** ***********
* ****** ********* ** ********* **** detection **** ****** ***********; **** ********* finds ***** *** *********** ********** *** the **** ******* **. ***** **** detection ** ***********, ** ** **** easier **** ***********. **** ********* ******* generally ***** *** ********* * ****, rather **** *********** * ***-**** ****** as * ****. ****** *********** ******** may ****** ** *** ***** ****** being ********** **** ************* * *****.
***** **** *** ********* *********, ****** recognition ******** ****** ** **** ********* performance, ** ****** *********** ****** ****** until * **** ** ********.
Vehicle *********
********* **** ******** ***** *** ***** an **** ** ** ******** *** alerting ** ************* ******* ** ******** subjects ** ****** ********* **.
******* ********* ** ************* ****** **** person ** **** ********* *** **** not ****** **** **** ** *** same **********.
Vehicle ********* **********
******** *** ***** ********** *** **** more ****** ** ******* ***** ****** detection **** ***** ****** ** ****** people ******* ******** ** ********:
*** **** ****** ********* *** ******* detection ** **** ******** *** ***** parked *** ***** ** * ****, and **** ********* **** ******** ******** alerts *** *** **** *******, ***** quickly ****** * ******** *** **********:
** *** ****** *******, **** ** a ****** *******.
Vehicle **************
******* **** **** ** ******* (*.*. car, *****, ***) *** ******** ** valuable *********** *** **************, *** **** AI-based ******* ********* ********* ******* ****:
*******, *** ***** ** ******** ********* vary, *** ***************** ** ****** *** SUVs, ********, *** *********** *** ******.
***** ******* **** ** ********** **** to *********, ***** ** ******* ***** and ****, **** *** ***** ********* are **** *** ******* ************ ******* detection *********. **** *** **** ****** with **** ******** ***/**** ********* *** require **** ****** ****** ** ****** the *** ************'* *****/****.