Projects

>The Smart Fence

Conventional security systems have limitations. For example, in an industrial setting, fence monitoring using video surveillance becomes expensive and inefficient as the area under observation grows to cover several miles in length. The goal of this project is to provide a sensor network approach to this problem. Using WirelessHART compliant communication nodes and cheap MEMS-based accelerometers, the solution covers recording vibrations locally on each fence section and relaying the readings back to a central computer for processing and detection. For the detection algorithm, decentralized sequential hypothesis testing was selected because it performed well both during testing and real deployments. In essence, the sensors communicate only when a decision is made concerning the observed event. Those decisions are then fused at the sink which raises the alarm or not. A long term deployment proved the applicability of this algorithm with 100% detection rate and 0% false alarms.

Following the successful showcase of the Smart Fence technology, this project aims at using MEMS and Electro-Chemical Sensors in combination with Low-Power radios to implement industrial wireless sensing applications. In particular, and at the Chevron-Richmond refinery, fence-line gas sensing is added to the previous security application with regular reporting of H2S, CO and VOC concentrations. Using MEMS accelerometers and magnetometers, valve position monitoring and machine vibration sensing are added for safeguarding both personnel and equipment. This project is concerned both with the COTS-based hardware and software behind each application.

Being part of the WSN group at UC Berkeley, I am also a contributor to the open-source initiative taken by Thomas Watteyne to publish a standards-based protocol stack for Wireless Sensor Networks. The source code, documentation and everything you need to know about this type of networks can be found at http://openwsn.berkeley.edu.

Tired with all the speculations around wireless sensor networks, we decided it was the time to demonstrate it was actually possible to transmit video over a multi-hop low power network. Not only did we succeed, but we did all of this using a standardized protocol stack. We summarized the results in an IEEE Comsoc E-letter that you can find in my publications page.

The US-DOT started a program to develop Intersection Decision Systems to address accidents that take place at those facilities. The UC Berkeley transportation department (under CEE), along with Caltrans and CA-PATH took up the "Signalized Left Turn Assist" aspect of this initiative. In broad terms, what we would like to do is sense whether we have people waiting to make a left turn at the intersection, then tell those motorists whether or not they have a big enough gap in the oncoming traffic to complete the maneuver safely. Of course, magnetometers and TSCH (time-synchronized channel hopping) came to the rescue. The study is reported in a paper that you can find also in my publications page.