Research Projects
SPECIAL MANPOWER DEVELOPMENT PROGRAM FOR CHIP TO SYSTEM DESIGN (SMDP - C2SD) (Ministry of Electronics and Information Technology )
MEITY Sanction No. and Date : 9(1)/2014-MDD, Dated:15.12.2014
Two-fold objective under SMDP-C2SD project (i) Group Project (ii) Individual Project
(i) Joint Project with IIT Roorkee cluster:
Broad objective of the joint proposal: Development of a working prototype of System on Chip for Remote Detection of Humans Trapped Under Debris in Disaster Affected Areas Using RF Sensing of Cardiopulmonary Motion. NIT Uttarakhand Team proposed and implemented FPGA subsystem implementation of the digital processing unit.
(ii) Individual Project
To develop "FPGA board level implementation of multilevel authentication electronic locker security system".
2. A framework for deriving context awareness of mobile or wearable device user (STMicroelectronics Inc, Santa Clara, CA, USA at IIT Delhi)
MEMS sensors such as accelerometer, gyroscope, magnetometer and barometer, in addition to microphone and camera are being widely incorporated in mobile phones, tablets and wearable devices due to their several utilities, small footprint and low cost. The inclusion of sensors in mobile and wearable devices has made it feasible to make the device context aware.
A probabilistic framework for base level and meta level context awareness about a mobile or wearable device user was presented that included continuous and transient motion and voice activities of the user, the spatial environment around the user. Each attribute of base level context awareness was modelled as a discrete-time discrete space random process. The time sequence of posterior probabilities of these random processes was called posteriorgram. The base level context awareness was represented by three posteriorgrams, for each continuous motion activities, voice activities and spatial environment around the device user, i.e., Motion Activity Posteriorgram, Voice Activity Posteriorgram, Spatial Environment Posteriorgram. The usefulness of the proposed framework was demonstrated by implementing descriptors of base level context awareness. The data for implementing the descriptors of base level context awareness was obtained using accelerometer, barometer, gyroscope, magnetometer and microphone. The probabilistic output of the framework for context awareness of mobile and wearable device user was indicative of the uncertainty of the output of each frame.
Outcomes: 3 patents granted, 1 patent in advanced stage of being granted and 1 journal publication.
3. Deep learning model based computer vision based rehabilitation exercises (Along with IT Gopeshwar)
Physiotherapy exercises like extension, flexion and rotation are absolute necessity for patients of cerebral palsy, traumatic injury, paralytic injury, post trauma stiffness, congenital deformity, spinal cord injury and Guillain Barre syndrome. A physiotherapist uses many techniques to restore movements needs in daily life including nerve re-education, task training, muscle strengthening and uses various assistive techniques. But, a physiotherapist guiding the physiotherapy exercises to a patient is a time consuming, tedious and costly affair. In the project, an automated system is designed for detecting and recognizing physiotherapy exercises using RGB-Depth camera that could guide the patients to perform the real time physiotherapy exercises without human intervention. Hybrid deep learning approaches are exploited for highly accurate and robust system for recognizing physiotherapy exercises of upper limb.
Outcomes: 3 journals in advanced stage of acceptance.
4. A signal processing algorithm for passive detection of open space or enclosed spatial environment was implemented. (STMicroelectronics Inc, Santa Clara, CA, USA at IIT Delhi)
Mobile phones and wearable devices are equipped with inbuilt high quality digital microphone or array of microphones that act as input for transmitting voice for communication or act as voice input for giving instructions to a specific application or personal assistant in the mobile or wearable device. Mobile and wearable devices are also equipped with processors that have the capability to perform complex computations, thus enabling the capability of knowing open or enclosed spatial environment around the device without explicit information from the device user. The accurate classification of open or enclosed spatial environment of the mobile or wearable device user is a useful contextual input for various context aware applications.
The open vs enclosed spatial environment of the mobile or wearable device user is detected and classified by estimating Spatial Environment Impulse Response (SEIR) and extracting novel features from it. The SEIR is estimated from the ambient sound signals received in a microphone or array of microphones without explicitly providing any known test signal. The features from SEIR are augmented with other features such as Mel-Frequency Cepstral Coefficients (MFCCs), delta MFCCs and double delta MFCCs from the microphone signal. The above features are provided as input to a pattern classifier such as a deep learning architecture for open vs enclosed spatial environment classification.
Outcomes: 1 patent filed and published (USPTO and European patent office)