Smart monitoring on airplanes
Nguyen, Thuy Diem.
Date of Issue2009
School of Computer Engineering
Parallel and Distributed Computing Centre
There is an increasing need for security surveillance systems in airplane ca-bins due to the threat of aircraft hijacking in recent years. Tackling this issue, this project aims to build an in-flight surveillance system called Smart Moni-toring System to monitor the location of all passengers during a flight. To carry out this task, the system uses a network of camera sensors inside the cabin. Each sensor node, which is placed in front of each passenger seat, consists of a processing unit and an infrared camera attached to it. The sensor nodes are connected to each other via a network for in-flight entertainment to form a vision sensor network. The processing unit of a node is required to perform several computer vision tasks in order to obtain useful information, in this case the passenger location, from the video data acquired by the camera. The computational cost required for processing those techniques is normally expensive. This is a sub-ject of concern for any system in an airplane cabin where both processing power of sensor nodes and electric power are limited resources. For that reason, the computer vision techniques chosen in this project are those which can accomplish the tasks and at the same time reduce the computational burden on the processing unit of each sensor node. A surveillance application called Smart Monitoring Application is respon-sible for processing data at each sensor node. Three main techniques which are employed for detecting, recognizing and tracking passengers’ locations are face detection, face recognition and a Hidden Markov Models (HMM) classifier to output the seat location of a passenger. Face detection is accomplished by utilizing a method for object detection called the Haar-like object detector. The training of a frontal face classifier is carried out using the FERET face database. The frontal face classifier gives a hit rate of about 90% and 15% false alarm rate on CMU-MIT frontal test set. Face recognition is done by using Principal Component Analysis (PCA) with K-Nearest Neighbour algorithm (KNN). The testing of the face recognition part gives an average performance of 92% on ORL face database. Three main challenges which can affect the performance of the detection and recognition algorithms are the variation in head poses of a passenger, different types of occlusions to the face and the varying lighting conditions in an airplane cabin during long flights. To address these issues, three additional vision techniques, skin detection, foreground detection and motion detection are employed. The outputs from all the detection and recognition algorithms will then be used as feature vectors for the HMM classifier which consists of a bank of HMMs. This classifier is utilized to compensate the error rate of each individual algorithm and output the seat location of a passenger. Furthermore, a combination of infrared cameras and suitable choice of lighting-variant algorithms can help to deal with the changing lighting condition issue. Although it has been implemented, the final integration of the HMM classifier into the system is still under testing. The results of this will be discussed during the project presentation.
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
Final Year Project (FYP)
Nanyang Technological University