|
Compensating surveillance camera movements using sequential image registration for car detection
|
Ali Karami *   , Mohsen Seriani , Masoud Varshosaz |
| K.N.Toosi University of Technology |
|
|
Abstract: (3424 Views) |
| Recognizing and detecting cars in videos is ones of the main issues in computer visions. The main assumption of surveillance cameras is that the camera is fixed while taking videos. Any movements due to wind or external forces causes frequent vibrations in the camera and shift of image pixels. This makes conventional methods to detect static objects as moving ones mistakenly. The main aim of this research is to register sequential frames with background image to remove the noise and the mentioned problem. The algorithm first extracts sequential frames and background image using median method, and then registers all frames with the background image. Cars are detected by subtracting background image from frames. Three data were used to evaluate the capability of the proposed methods. Each data set has different car density and contain frequent vibrations while recording the videos. To evaluate the proposed method, FAR, MODP, HR and error percentage criteria were calculated with and without registration. The overall accuracies were 89% and 76% respectively which shows the 13% improvement in the accuracy of detecting cars with registration technique. |
|
| Keywords: registration, object recognition and detection, matching, moving objects, feature extraction |
|
|
Full-Text [PDF 1776 kb]
(1088 Downloads)
|
Type of Study: Research |
Subject:
Aerial Photogrammetry
Received: 2016/05/28 | Accepted: 2017/01/17 | Published: 2018/03/19
|
|
|
|
|
|
|
| Send email to the article author |
|
|
