Elements of a multi-object tracking systems

• vague 모호한

What makes multi-object Tracking difficult?

• prediction과 measure 겹치는 부분을 tracking 해서 멀티 옵젝트 tracking을 한다.

• 하지만 만약 아래와 같은 문제가 생긴다면 어떻게 할 것인가?

• this is data association problem

the number of objects being tracked is not fixed

• sometimes tracks need to be created or removed.

• 만약 prediction과 measurement가 멀다면 에어플레인의 track을 delete한다.
• 하지만 false positive나 missed detection이 된다면 큰 오류가 발생한다.
• this is track maintenance problem

when tracking multiple objects what are some ways to approach

• data association
• track maintenance

obesrvation step

• all those thing is handling in Observation steps

assignment step

• Matching an obesrvation to a track

• Eclidian distance로 assignement를 하는것이 아는 probability distribution으로 assignement를 한다.

track maintenance

• deleting and creating tracks

estimation filtering

• we may choose to ignore obseravtion outside of a defined region for each track

gating

• it’s a screening mechanism that determines which detections are valid candidate to look at for assignment and which should just be flat-out ignored for

Exercise

• GNN(Global nearest neighbor) for Data association

• IMM for estimation filter

• speed up

• JPDA for Data association