1. Action Localization based on Video Object Detection (Nov. 2013 -- June. 2014)
We propose a novel spatio-temporal filtering technique to improve the
per-pixel prediction map, by leveraging the spatio-temporal smoothness of the
video signal. Different from previous techniques that perform spatio-temporal filtering
in an offline/batch mode, e.g., through graphical model, our filtering can be
implemented online and in real-time, with provable lowest computational complexity.
Moreover, it is compatible to any image analysis module that can produce
per-pixel map of detection scores or multi-class prediction distributions. For each
pixel, our filtering finds the optimal spatio-temporal trajectory in the past frames
that has the maximum accumulated detection score. Pixels with small accumulated
detection score will be treated as false alarm thus suppressed. To demonstrate
the effectiveness of our online spatio-temporal filtering, we perform three
video event tasks: salient action discovery, walking pedestrian detection, and sports
event detection, all in an online/causal way. The experimental results on
the three datasets demonstrate the excellent performances of our filtering scheme
when compared with the state-of-the-art methods.
Efficient Online Spatio-Temporal Filtering for Video Event Detection
Xinchen Yan, Junsong Yuan, Hui Liang, Liqing Zhang
In ECCV 2014 Workshop on Video Event Categorization, Tagging and Retrieval towards Big Data (VECTaR 2014) [pdf][bib][slides][code]
Keywords:
Video Event Detection, Object Detection, Spatio-Temporal Modeling
Abstract:
2. Video Object Detection (July. 2013 -- March. 2014)
Other Projects:
* Brief Description: An offline-trained human detector using Implicit Shape Model (Hough Voting). Specifically, codebook is obtained by Hierarchical K-means clustering on image patches extracted on training images. Also, random forests is applied in order to improve localization accuracy. Results illustrate that method performs well in different object scales, poses, and complex backgrounds on Weizmann Dataset, INRIA Person Dataset, and UT-Interaction Dataset.
Visual Object Tracking (Jan. 2013)
Visualized detection results on UT-Interaction Dataset
* Brief Description: In this project, an object tracking system is built using Mean-shift algorithm and Kalman Filter. Color and texture histogram has been utilized for calculating patch distance.
Visualized tracking result on SJTU-Football Dataset
"SIFT feature" based Image Retrieval (Aug. 2012)
* Brief Description: A simple image retrieval system based on "SIFT feature" is developed. To retrieval images with efficiency, the weighted vocabulary tree based on TF-IDF mechanism is applied. As for visualization, the best k matching results are selected using nearest neighbour search (in this case, k = 5).
Auto Stitch (July. 2012)
Visualized retrieval results
* Brief Description: A photo stitch tool that calculates homography using RANSAC algorithm. Weighted average blending is utilized in the stitched results.
"MOPS feature" based Image Matching (July. 2012)
Visualized stitching results