Scribd
Upload
45 views4 pages

12 CS1AC16 Detection and Tracking

Object detection aims to automatically find objects in scenes. Detectors take images as input and output the position of detected objects with bounding boxes. Tracking follows detected objects to understand their movement trajectories over time. The simplest detection method is background subtraction, but it requires a clean background image without objects. Change detection compares frames to detect changes while avoiding issues from background variations. Feature-based detection locates objects by specific attributes like color. Tracking connects detections into trajectories using algorithms like distance minimization or the Hungarian method to match detections across frames. Tracking has applications in security, behavior analysis, scene monitoring, and healthcare.

Uploaded by

Anna Abcxyz
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
45 views4 pages

12 CS1AC16 Detection and Tracking

Object detection aims to automatically find objects in scenes. Detectors take images as input and output the position of detected objects with bounding boxes. Tracking follows detected objects to understand their movement trajectories over time. The simplest detection method is background subtraction, but it requires a clean background image without objects. Change detection compares frames to detect changes while avoiding issues from background variations. Feature-based detection locates objects by specific attributes like color. Tracking connects detections into trajectories using algorithms like distance minimization or the Hungarian method to match detections across frames. Tracking has applications in security, behavior analysis, scene monitoring, and healthcare.

Uploaded by

Anna Abcxyz
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

CS1AC16 DETECTION AND TRACKING

The aim of object detection is to get automated systems to automatically discover interesting objects
in the scene.
 Detected objects can then be used by other subsystems to do something interesting
 What's interesting depends on the system application
The detectors take images an inputs and generate target positions – a bounding box containing the
object

Finding objects of interest


 Moving objects are often more interesting than static – for example people and vehicles in
contrast to buildings
o However, this depends on the application – e.g. injured people might not move much,
whereas trees sway in the wind.
o The camera may also be moving (e.g. mounted on a vehicle), complicating the issue
further.
 Calibrating the scene can help detection (telling the system where particular reference point are
in the scene, e.g. the ground).

The simplest way of detection is background subtraction:


 The difference between the image and the background is where the objects are
 The problem is that we need a good reference frame for that – a picture of the background with
nothing in it. Furthermore, the light conditions change throughout the day, which may affect
detection.

Change detection
 Instead of using the first frame as reference, we can use the previous one – subtracting it from
the current frame we get an update on what has changed in the scene
 In this way we get continuous updates of what has changed in the picture – we do not need to
worry if the first frame was a good reference
 It's important to threshold the results – this can get rid of things like trees swaying in the wind.
Setting a wrong threshold can result in the loss of some objects of interest.

Feature based detection


Change detection and background subtraction were good if the target was moving and the camera
static. Feature based detection allows for detection of objects with a particular feature – for example a
particular color.
Histogram of oriented gradients (HOG) is a feature descriptor used for object detection. It involves
counting occurrences of gradient orientation in localized portions of the image. The theory behind that
is that local object appearance and shape within an image can be described by the distribution of
intensity gradients or edge directions.

Object tracking – following an object to see where it's going


Knowing about an object's movement allows us to:
 Make inferences about the object (is it behaving unusually or going where it's not allowed?)
 Make inferences about the scene (obstructions, paths of least-resistance)
 Keep track of the objects we already know about (is it the same person from 10 minutes ago or
a different one?)

Trackers take positional data and generate a trajectory – a description of movement through the scene
There are 2 types of trackers:
 Single target – following a single objects
 Multi target – following many objects without mixing them up
One of the simplest trackers is distance minimization tracker – for each target in a frame it matches a
target in a previous frame so that the total distance of movement is minimized.

Hungarian method – an algorithm for finding the total minimum distance between pairs
Set up: make a table of all possible distances between pairs:

Step 1: For each row, find the minimum value and subtract it from every element in that row (if
each column contains exactly one 0, finish at this step):

Step 2: For each column find the smallest value and subtract it from all elements in the column
(if each row contains exactly one 0, finish):
Step 3: Cover the grid with the minimum number of lines so that all 0s are covered (if the
number of lines equals to the number of rows/columns, stop):

Step 4: Find the lowest value from all uncovered elements. Subtract that value from all
uncovered elements. Add this value to every element covered by 2 lines.

Repeat step 3 and 4 as needed.


A 0 in final table represents the assignment of rows to columns.
Tracking is used for:
 Security – UK is one of the most surveyed countries
o Critical infrastructure monitoring
o Critical assets monitoring
o Maritime mobile assets monitoring
 Identifying suspicious behavior
o Protecting cargo ships
o Infrared cameras are often used for monitoring in the dark
 Scene flow management – effectively managing a lot of moving objects
 Sports – objects travelling at high speed
o Aiding referees and minimizing errors
 Healthcare – looking after elderly people in their homes – minimizes the need for home visits
 Monitoring in retail

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy