Video sensors (video cameras) and wireless broadband networks are becoming ubiquitous features of modern life. The confluence of these two technologies now makes it possible to construct wireless ad hoc networks of multiple video sensors that can be rapidly deployed, dynamically configured, and continuously operated to provide highly-available coverage for environment monitoring and security surveillance. While many extended "eyes" are being installed at an unprecedented pace, the intelligence needed for interpreting video-sensing events by computers is still rather unsophisticated. In addition, algorithms that can scale well with the number of sensors and high-volume of data are yet to be developed for effectively managing and mining video-sensor data streams. In order to develop a "brain" behind a large number of optical "eyes" to support (semi-) automatic event sensing, statistical methods are essential for improving the two major phases of event sensing: 1) data fusion and 2) event mining. The data-fusion phase integrates multi-source data to detect and extract motion trajectories from video sources. The event-mining phase deals with classifying the events as to relevance for the query. This talk presents recently developed statistical methods in energy-preserving data sampling and filtering, spatio-temporal data fusion, sequence-data mining, change and outlier detection, and multi-resolution and multimodal statistical learning. We will also discuss future directions that research might take.