motion

Motion is a variation in the observed scene due to objects or subjects that are changing their physical position. This is one of the basic tasks with natural applications in many fields, such as traffic monitoring, video surveillance, people tracking and behavior analysis. Traditional approaches include background subtraction, frame by frame difference, optical flow estimation. The separation between motion of the subject movement, i.e. a drone or a robot, and motion of framed multiple objects is considered a difficult problem due to motion noise, image blurring, presence of repetitive uninteresting motion, illumination variations, and changes in the objects' appearance.

MATCHING VIEWS OF MOVING CAMERAS

Moving cameras, such as on collaborative robotic platforms, need to recognize the same place observed through the time from different viewpoints. Matching views is challenging for independently moving cameras that directly interact with each other. The temporal changes of a 3D point at multiple scales have to be captured. We proposed a multi-scale binary descriptor based on ORB that improves the accuracy of feature matching under scale changes. MORB describes an image patch at different scales using a properly oriented sampling pattern of intensity comparisons in predefined pixel pairs. For this multi-scale descriptor, we also propose a matching strategy that estimates the cross-scale match between MORB descriptors in different views.


related publications

A. Xompero. Local features for view matching across independently moving cameras, PhD Thesis, Queen Mary University of London, UK, 2020, supervisors Andrea Cavallaro (QMUL) and Oswald Lanz (FBK)

A Xompero, O Lanz, A Cavallaro. A spatio-temporal multi-scale binary descriptor, IEEE Transactions on Image Processing, Vol. 29, pp. 4362-4375, 2020

A Xompero, O Lanz and A Cavallaro. MORB: A Multi-Scale Binary Descriptor. IEEE International Conference on Image Processing - ICIP, pp. 2167-2171, 2018

LOW POWER CAMERAS

Motion is a typical feature in event detection for surveillance purposes, in order to capture objects that change their position over time. To be embedded in battery power cameras algorithms of motion detection must be simple but reliable. A double-threshold dynamic background subtraction algorithm has been proposed for the real-time detection of moving objects in gray-level videos depicting indoor and outdoor scenarios. The background is continuously modeled at each pixel by two thresholds that update over time. A pixel is labeled as a motion pixel if its intensity value falls out of the range bounded by the two thresholds. The resulting motion map is de-noised through programmable morphological filters. When the sensor detects sufficiently large regions of motion, it generates an alarm that is sent to the external processor for further actions.


related publications

Y. Zou, M. Gottardi, M. Lecca and M. Perenzoni. A Low-Power VGA Vision Sensor With Embedded Event Detection for Outdoor Edge Applications, IEEE Journal of Solid-State Circuits, 55(11):3112-3121, 2020

Y. Zou, M. Lecca, M. Gottardi, G. Urlini, N. Vretos, L. Gymnopoulos and P. Daras. A Battery Powered Vision Sensor for Forensic Evidence Gathering, International Conference on Distributed Smart Cameras - ICDSC, 2019

BACKGROUND UPDATING

In the case of a static camera the technique of background subtraction can be used to detect and segment moving objects in a dynamic scene. It consists of maintaining a background image of the scene then detecting foreground objects by subtracting it from the current frame. Background updating is a critical task particularly in outdoor scenes which undergo significant changes caused both by natural events, e.g. the sun suddenly disappearing behind clouds, or artificial events, like the change of the exposure time of the acquisition device, or the switching-on of artificial lights. A background updating module should be able to deal with gradual and sharp illumination changes. We proposed an algorithm based on a Kalman filter that takes into account such phenomena by measuring the global illumination change and using it as an external control of the filter. This allows the system to better fit the assumptions about the process to be modeled. Furthermore, to improve the accuracy and robustness of the algorithm, we included a method to deal with the problem of saturated pixels.


related publication

S. Messelodi, C.M. Modena, N. Segata and M. Zanin. A Kalman filter based background updating algorithm robust to sharp illumination changes, 13th International Conference on Image Analysis and Processing - ICIAP, pp. 163-170, 2005