This document discusses object detection and tracking algorithms. It covers the following key points: - Background modeling uses a mixture of Gaussian model to identify foreground objects from backgrounds. Tracking includes correlating current objects to previous ones and handling occlusion. - Challenges include noisy images with false positives, shadows, and removing noise while detecting moving regions. Distance filtering and morphological operations help address these. - Objects are identified using contours and bounding boxes. Kalman filtering can be used for prediction but has difficulties with smoothing. - Performance is evaluated using multiple object tracking accuracy (MOTA) and multiple object tracking precision (MOTP) metrics. Areas for improvement include handling size changes and removing lost objects.

2.  Object
 Frame
 Framelist

3.  Background Modeling – Gustav
 Foreground Processing – Martin
 Object Identification – Mattias
 Prediction and Evaluation – Alexander

4.  Uses a mixture of Gaussian model described
by Wood.
 Update procedure is slow... Close to 1 second
per update on a larger image.

5.  Noisy, lots of false positives.
 False positives are mostly isolated.
 Easy to handle with later processing steps.

6.  Three main objectives:
 Suppress shadows
 Remove noise
 Detect moving regions

7.  Algorithm implemented as described in the
master thesis by John Wood.
 HSV mapping:
 Easy to implement
 Good performance
 Few false positives
 Problems with gray areas

9.  “Distance filtering”
 Throw away foreground regions not thick enough
 Good performance
 Slow?
 Implementation
 cv::findContours, cv::pointPolygonTest
 Iterate over bounding rectangle
 Measure distance inside contour only
 Final touch: some morphological dilate

10.  Object creation
 Find remaining contours
 Create bounding boxes
 Calculate positions

11.  Uses cv::findContours, cv::boundingRect
 Find outer contours
 Create boundingrect for each contour
 Use the bounding rectangle to add objects to
the frame’s object list.

12.  Objectives
 Correlate previous objects with current objects
 Handle occlusion
▪ Objects <-> Objects
▪ Objects <-> Background
 Assign unique ID’s to new objects
 Forget objects that leave the screen

13.  A measure of how similar two objects are.
 A measure of how probable it is that two
objects are the same.
 휖 = 푥1 − 푥2 − 휕푥2
2 + 푦1 − 푦2 − 휕푦2
2 + 푤푖푑푡ℎ1 − 푤푖푑ℎ푡2 + |ℎ푒푖푔ℎ푡1 − ℎ푒푖푔ℎ푡2| 2
Squared euclidian distance Squared size difference
 Discards outliers

14. Overlapping move Non-overlapping move
1 2
Parent split
Sibling merge
4
Lost Discovered
6
3
5

16.  Assumes all passed objects are ”real”
 Large objects tends to collect lost heads, feets...
 Width and Height should not change too fast...
 The error function isn’t tuned at all: a change in
width,height should probably impact more.
 Objects should be removed if they have been lost
for too long. Use the variance estimate from the
kalman filter?

17.  Kalman: The optimal linear predictor
 Components
 State-Space Model
 Covariance Matrices
 Difficulties
 Smoothing

19.  MOTA & MOTP
 Easy to understand
 푀푂푇퐴 =
푚푖푠푠푒푠+푓푎푙푠푒푃표푠푖푡푖푣푒+푚푖푠푚푎푡푐ℎ푒푠
표푏푗푒푐푡푠
 푀푂푇푃 =
푑푖푠푡푎푛푐푒
푚푎푡푐ℎ푒푠

20.  Improvement
 No area evaluation
 Get rid of the threshold