Work


RESEARCH WORK

Project Associate, Robert Bosch Centre for Cyber Physical Systems, IISc Bangalore

Oct 2017 - Present
Industrial Internet of Things
Supervisor: Prof. Rajesh Sundaresan
  1. Industrial IoT Stack Implementation Details (Github)

  2. Network Layout (Github)


Project Associate, Electrical and Communication Engineering, IISc Bangalore

August 2016 – September 2017
WSN in detection and early warning system for wildlife monitoring and protection
Supervisor: Prof. P Vijay Kumar

Worked on a low power sensor platform which uses Passive Infrared sensor array for intrusion detection and classification between human and animals like big cats. The sensor platform had to be low power and would be running from a solar energy harvesting system. Correlation between a pair of PIR sensors and the signal strength from 4 other PIR sensors were used as a feature vector and fed to an SVM running on a low power mote (Wismote and Zolertia Re-mote) to classify the intrusion as human or animal. Also developed light weight image processing techniques for Raspberry Pi Zero to complement the above platform at times of day when temperature gets close to body temperature of humans or animals, because PIR performs poorly in such environments. The work involved extracting features from video feed which are then fed to an SVM classifier to classify between the Human, Animal and vegetation. We achieved a frame rate of 8 fps on Raspberry Pi-Zero, 25 fps on an Orange Pi-Zero using an optimized version of the algorithm which provided accuracies of about 93%. An Odroid C2 was also deployed whose frame rate was 25fps with a little heavier version of the algorithm for better accuracies reaching 97%. The work also involved comparison of various permutations of different image processing/feature extraction techniques like optical flow and machine learning models like decision trees, SVM and K- Nearest Neighbour based on their computational and energy efficiency while running on different Single Board Computers (SBCs) available like RPi2, RPi3, Odroid C2 and Orange Pi.

My role in the project spanned many areas including hardware, embedded systems, WSN, computer networking, image processing, signal processing and machine learning. On hardware end, I designing the signal conditioning multistage filter and amplifier for the PIR sensors and selecting the right ADCs. On the embedded aspect, my contribution was towards ADC interfacing of these sensors via I2C to a mote running Contiki-OS (Re-Mote). I also contributed in part to develop routines for extracting the energy and correlation features from the PIR signal sampled via the ADC in the ContikiOS. On the WSN side, I helped implement UDP client codes on ContikiOS for sending decisions from the Re-Mote to application server. I was completely responsible for developing the complementary camera platform. A few of my tasks were to implement optical flow using OpenCV, build various features from the optical flow vectors and implementing an SVM classifier which uses these features. Also included were tasks for optimizing the above systems for running on SBCs to obtain higher frame rates which lead to understanding of systems programming, IO bound vs CPU bound tasks and later coming up with a pipelined architecture for the optical camera using multithreading to address camera IO bottlenecks and comparison of effects of using various image filters on computation and accuracy of classification on these SBCs.

Papers:
  • Tarun Choubisa, Sampad B. Mohanty, Kodur Krishna Chaitanya, Mohan Kashyap, Sridhar A, Akshay Singh P. Vijay Kumar, Fellow, IEEE, “A reduced-complexity, reduced-power camera system for intrusion classification in an outdoor setting”, VisionNet - ICACCI, Manipal, 2017 [BEST PAPER] [PDF]

  • Tarun Choubisa, Mohan Kashyap, RN Rithesh, Sampad B Mohanty, “Direction and gender classification using convolutional neural network for side-view images captured from a monitored trail”, IEEE ICIIP, Shimla 2017 [BEST PAPER] [PDF]

  • Tarun Choubisa, Sampad B. Mohanty, Mohan Kashyap, Kodur Krishna Chaitanya, Sridhar A, P. Vijay Kumar, Fellow, IEEE, “LITE: Light-based Intrusion deTection systEm Using an Optical-Camera and a Single Board Computer”, LCN Demos Track, IEEE LCN, Singapore, 2017 [PDF]

  • Tarun Choubisa, Mohan Kashyap, Sampad B. Mohanty, P. Vijay Kumar, Fellow, IEEE, “Comparing chirplet-based classification with alternate feature-extraction approaches for outdoor intrusion detection using a PIR sensor platform”, ICACCI, 2017 [PDF]

  • Tarun Choubisa, Sampad B Mohanty, Mohan Kashyap, Shivangi Gambhir, Kodur Krishna Chaitanya, A Sridhar, P Vijay Kumar, “An Optical-Camera Complement to a PIR Sensor Array for Intrusion Detection and Classification in an Outdoor Environment”, SenseApp-IEEE LCN, Singapore, 2017 [PDF]

Keywords: OpenCV, Optical Flow, Image Processing on Multicore, Contiki-OS, Correlation, Background Subtraction,Support Vector Machine

Project Assistant, Electronics and Communication Engineering, NIT Rourkela

June 2016 -July 2016
Embedded System for Mine Safety and Heavy Earth Moving Vehicle Tracking
Supervisor: Prof. Debiprasad P Acharya

In this period I worked on deploying the system I developed in my M.Tech project “Vehicle Monitoring in Open Cast Mines under Unreliable Cellular Network using ZigBee”. I also helped in the technology and skill transfer to the staffs of the funding company - Jindal Steel and Power - Tensa.

More details in my work as Research Assistant in ECE Dept. NIT Rourkela in the year 2015-2016


Research Assistant, Electronics and Communication Engineering, NIT Rourkela

June 2016 -July 2016
Embedded System for Mine Safety and Heavy Earth Moving Vehicle Tracking
Supervisor: Prof. Debiprasad P Acharya

Funded by Jindal Steel and Power Limited, the project aimed to tackle the problem of near real-time vehicle tracking in an open cast mine which suffers from a poor GSM network. I developed both the hardware and software for the system to solve this problem using a ZigBee network that collects the position and speed data from vehicles when they are not under GSM coverage and then relays it to a ZigBee node which has access to GSM network and hence uploads the acquired data to the Vehicle Monitoring Server over GPRS. The codes were developed using Python for a Raspberry Pi computer which acts as the gateway node, in C++ for the vehicle tracking equipment (mounted on each dumper/vehicle to be tracked) and again Python, HTML and JavaScript for the Vehicle Monitoring Server. The system was deployed at Jindal Iron Mines at Tensa, Sundergarh Odisha India and has been working since March 2016. The project details can be found in my M.Tech thesis which can be accessed at [Thesis.pdf] Project Website (Authentication Required – Guest credentials for demo available on request) : nitrjindal.in

Papers:
  • Himansu Sekhar Pradhan, Sampad Bhusan Mohanty, Santosh Madhukar Yerme, Paresh Govind Kale, Debiprasad Priyabrata Acharya, “Embedded System for mine process monitoring in a network constrained environment using wireless communication bridge”, ICMOCE, IIT Bhubaneswar, 2015 [PDF]

Founding Member and Lead Embedded and Network Developer at Pheonix Robotix

PheonixRobotix
AURASSURE

One of the first team member of the startup Pheonix Robotix (spin off from the Balloon Satellite Lab at NIT Rourkela and the Air Quality Sensor Network team) where I worked as the lead embedded developer after class hours. I managed a team of 3 developers to develop the software stack for an Air Quality Monitoring Kit called AURASURRE (https://aurassure.com). I also led the networking of the devices with the State Pollution Control Boards (SPCB, Govt. of Odisha, India) pollution monitoring servers during which I visited and interacted with the SPCB technical staff to develop an API for the same. My work also included writing driver codes on Arduino for various air quality sensors including sensors from the well-known company Alphasense.

Aurassure and TraDe
ONLINE POLLUTION MONITORING OF SOFT IRON MANUFACTURING PLANTS (LATER CALLED PRODUCT “TRADE”)

I was responsible for developing embedded code on Arduino platform for interfacing a hardware device we developed at PhoenixRobotix which collected pollution data like SO 2 and particulate matter from the Electrostatic Precipitator and uploaded it to the Orissa State Pollution Control Boards Server over GPRS for continuous monitoring of the pollution. All transmissions were encrypted using AES encryption and a custom application packet was used to be sent over raw TCP. HTTP was not used deliberately to decrease data usage due to header overheads. The device used industry standard 4ma-20ma current loop for collection of data from the sensors. Finally a generic device named TRADE was developed which could be easily deployed in any similar factory and start transmitting data without much manual configuration.