Motion

One obvious kind of processing one may want to do with a video sequence is the detection of moving objects in the scene. We discuss how moving objects can be detected. This is not meant to replace your reading or the lecture

Detecting motion implies examining the spatiotemporal variation of pixel values in the scene.

spatiotemporal = spatio refers to the 2D spatial domain of an image in the video sequence. temporal refers to the 1D time-line of the successive images/video frames in the video.

Motion Field

This is a diagram containing motion vectors that indicate the direction and speed of the moving portions of image. Below is an image with the a set of superimposed vectors on top of it. Note that the Car is moving down the road to the left.

This was derived from the following sequence:

Frame 0

Frame 1

Frame 2

Uses of Motion Detection

Video Compression.
Target Tracking.
Speed Calculation.
Depth Perception.
Security Systems.

Assumptions

Some of the differences in algorithms used to detect motion arrise from the different assumptions about the world that are made. Here are some of them:

Image Intensity will be preserved.
- The position of a particular intensity will change between images and this is how we measure motion.
- This resulting motion field is refered to as intensity flow
Edges will be preserved
- The position of a particular edge will change between images and this is how we measure motion.
- This resulting motion field is refered to as edge flow
Whether or not the camera is fixed.
If the camera is moving, whether or not objects in the scene are moving.

Commercial System

Difficulty: Finding the Correspondence:

What should correspond: Objects, features or intensities (pixels)?

A curiosity: Motion Detection Sensor

A good example of a low-performance application is an infrared motion detector. Nowadays, it is common to offer backyard lighting systems or door opening systems which detect the presence of a moving object with elevated infrared emission. If you wave your hand about, the infrared scene that can be detected features a variation in the infrared signal to some pixel of an imaging system. So what is needed is an array of detector elements and some sort of focused optics. Without the focused optics, moving your hand about does not produce a change in the total illumination- and would not produce a variable signal. Remember that the pyroelectric detectors do not detect heat - only changes in heat.

So, it has become common to package a PVDF detector array in a low-cost optical package which uses a Teflon lens to focus the light. Teflon lens material is also inexpensive, and is transmissive enough in the IR that is does a decent job.

Typical Teflon lenses used in motion detection systems are made with a surface texture that includes several circular bumps. These bumps act as focusing lenses, and will bring light from a particular part of the scene to the detector. As a warm object moves through the scene, radiation is occasionally focused on the detector, causing a transient in signal which is detected.

Some Fun Links

Depth Perception from Motion

One Probaby Brain Circuit for Motion Detection

Motion Detection using Hough Transform

Links on Tracking

Paper on Tracking in Video Surveillance

Paper on Real-time Tracking

Paper on Tracking multiple objects (humans)