I first heard about computer vision in high school. The US TV drama “Person of Interest”, which flaunted the formidable identification …show more content…
I distinguish myself especially in the specialized courses. For example, I achieved the highest point in many computer-majored courses, such as operate system and compiler theory. Such accomplishments implied persistent endeavor, active thinking, and analytical mindset, attributes which are sure to pay off when undertaking studies in computer science. I can also keep pace with the new developments in technology through extensive reading, develop comprehensive set of capabilities in the course projects and become an independent thinker through class discussions. Believing “Experience is the best teacher”, unlike other girls who lack of the ability of practice, I am totally in to it. I can write concise code, debug it, grab the key factor of the problem and be the first one who finish the experiments. To be even better equipped for future study, I seized every chance to participate in professor’s speech and communicate with scholars. Upon graduation, I will surely be prepared for any academic hardship that lies …show more content…
With the deeper study on these specific topic and thoroughly discussions with professor and graduate student, I became interested in deep learning and the wide use of surveillance camera drove me determine to focus action recognition. Deep learning, which aims to achieve intelligence in machines, and the field of Neural Networks, which mimic the properties of biological neurons, are powerful tools which can make computer think as human, recognize the movement and respond promptly. After weeks reading international conference and top publications’ paper about this area, I have a basic understanding of it and want to make some improvement based on the experts’ work. Since the recent work on action recognition is about a part of body and use only one layer of CNN (Convolutional Neural Networks), I come up with a new idea that I can add more layers in order to gather more sequence information and other details since some movements are too similar for computer to recognize. Besides, dividing body into five parts is a convenient solution, but not perfect. Some small range of movement might be neglect and the relationship between joints will not be found. Since the movement of people are various and successive, we should pay more attention to these feature. I am now doing model training in Metconvnet and Caffe (Convolution Architecture For Feature Extraction) and try to build the best net to do it. This