Motion Based Object Detection and Classification for Night Surveillance

This paper describes a simple technique for object detection and temporal data association of thermal image sequences. Night surveillance system using thermal imaging involves object detection, temporal data association and tracking of object. Object detection could be motion based or feature based. The temporal data association in multi-object classification involves finding the minimum distances between an object in current frame to the objects in previous frame. A performance comparison is made between two techniques for object detection based on timing constraints and qualitative analysis. The second method proposed clearly outperforms the first in terms of timing. Target classification using neural network dynamically identifies the moving object. Implementation is done using MATLAB software.