In this thesis an intruder detection system is developed using an omnidirectional camera. The omnidirectional camera is constructed using low-cost off-the-shelf components, and has a wide view employing a hemispheric mirror. An indoor security robot i...
In this thesis an intruder detection system is developed using an omnidirectional camera. The omnidirectional camera is constructed using low-cost off-the-shelf components, and has a wide view employing a hemispheric mirror. An indoor security robot is a particularly considered application in the system development. The camera system is installed on the top of a mobile robot. To detect intruders from the output images of the omnidirectional camera, motion regions on the images are extracted by an adaptive background subtraction technique. The owner and an intruder are distinguished by checking the keyboard input of a preset password to a host PC. Once the system decides that a person detected is an intruder, a warning text message is sent to the owner. In addition, a series of scene images are uploaded to a webpage so that the owner can check the images at a remote site. While an image of the omnidirectional camera can contain wide area, it is extremely distorted. Therefore, the images uploaded on the webpage are transformed to panoramic images so that the user can understand the scene more easily. The system is designed to run steadily in various surroundings and situations. Image characteristics are analysed particularly to avoid malfunctions due to sudden indoor lighting change. The system developed worked well in experiments done with different numbers of target people at various distances.