In this project, we design a real-time human-computer interaction system based on hand gesture. The whole consists of three components: detection, gesture recognition and (HCI) recognition; realizes the robust control mouse keyboard events with higher accuracy recognition. Specifically, use convolutional neural network (CNN) to recognize gestures makes it attainable identify relatively complex using only one cheap monocular camera. We introduce Kalman filter estimate position which cursor is...

In this paper, we propose a novel approach of video text keyframe detection, to achieve the goal representing with textual keyframes and reducing waste resources in review. Different from works summarization which mainly focus on variation scenes videos, pay attention variances between sequential frames. For above purpose, Text Siamese Network (TSN) is developed automatically detect contain videos. Specifically, TSN composed two branches, similarity measurement identification. The first...

A method to detect generic objects by training with a few image samples is proposed. new feature, namely locally adaptive steering (LAS), proposed represent local principal gradient orientation information. voting space then constructed in terms of cells that query coordinates and ranges feature values at corresponding pixel positions. Cell sizes are trained spaces estimate the tolerance object appearance each location. After that, two detection steps adopted locate instances class given...

Shape template matching is an important approach in object detection and recognition.In this paper, we propose a fast novel method, which represents edge map contours with salient points retrieves the target by using backtracking method two stages from coarse to fine matching.Our has main contributions.One way represent contour structure points.The other that proposed can directly operate on real images, improves its practicability.According experimental results, speed than previous works.

The uncertainty and variability of underwater environment propose the request to control robots in real time dynamically, especially scenarios where human need work collaboratively field. However, imposes harsh restrictions on application typical communication methods. Considering that gestures are a natural efficient interactive way for human, we, utilizing convolution neural network, implement real-time gesture-based recognition system, who can recognize 50 kinds from images captured by...

One-stage object detection approach which utilizes multi-scale feature maps to predict objects is currently the best real-time detector. However, in this approach, high-resolution are responsible for detecting small harder learn a proper abstraction of than low-resolution maps. The problem that these have transform sufficient low-level information next layer while learning high-level abstraction. In paper, we develop transformation module adopts dense structure simplify addition, utilize...

Contour representation is an important application in image compression, template matching, object detection and recognition. However, it far from meeting the current requirement due to expensive computational cost complex noise real-world application. In order make contour more practical, we propose a novel approach of abstracting contours objects image. our approach, firstly find salient points on target by combining ellipse model Chord-to-point distance accumulation techniques. Then,...

Investigating that some face regions are possibly more reliable than the others when verifying two images due to local abnormal differences caused by uncontrolled factors in unconstraint environment,we propose a novel verification algorithm based on pairwise pre-estimation. In our algorithm, we estimate reliability of region detecting key facial points image pair. Then implement classifications such and combine results generate final recognition result. Furthermore, classification, also...

Shape template matching is an important approach in object detection and recognition.In this paper, we propose a fast novel method to represent edge maps by using salient line segments, which used detect objects based on shape matching.Firstly, image edges are computed and, these fragments, corner points extracted.Then, parabola model proposed edges.Secondly, points, directional chamfer framework compute the similarity between corresponding locations target image.Our has two main...

Chinese keyword spotting is a challenging task as there no visual blank for words. Different from English words which are split naturally by blanks, generally only semantic information. In this paper, we propose new spotter natural images, inspired Mask R-CNN. We to predict the masks guided text line detection. Firstly, proposals of lines generated Faster R-CNN; Then, and predicted segmentation in proposals. way, keywords parallel. create two datasets based on RCTW-17 ICPR MTWI2018 verify...

Chinese keyword spotting is a challenging task as there no visual blank for words. Different from English words which are split naturally by blanks, generally only semantic information. In this paper, we propose new spotter natural images, inspired Mask R-CNN. We to predict the masks guided text line detection. Firstly, proposals of lines generated Faster R-CNN;Then, and predicted segmentation in proposals. way, keywords parallel. create two datasets based on RCTW-17 ICPR MTWI2018 verify...