Interest point detection
Proposes a more appropriate treatment of the temporal dimension for detecting video interest points by using vector calculus on optical flow. Also, demonstrates that the proposed method is more robust to common video transformations in a video copy search setting-scale and compression ratio changes- when
compared to other methods.
Demonstrates that the proposed interest point detector gives improved action recognition performance on a popular benchmark dataset by combining it with local descriptors in a bag-of-features framework.
Person detection to control the AC in offices or home
Person detection was done using DNN, Opencv on Raspberry Pi. Temperature and Humidity sensor was used to get the current temperature and humidity. Based on these three output AC was switched on/off in different modes.
Facial emotion recognition
CNN based architecture is used for emotion recognition from videos. Keras with tensorflow backend and OpenCV is used for coding.
Person/object detection is done using CNN modelse.g. Fast RCNN.
face detection and recognition is done using hand-crafted features
Currently working on various computer vision projects such as facial emotion recognition, object detection, activity detection. Interested person can contact for collaboration or help.