In this paper, we improve upon the accuracy of existing tracklet generation methods by repairing tracklets based on their quality evaluation and detection propagation. Starting from object detections, we generate tracklets using three existing methods. Then we perform co-tracklet quality evaluation to score each tracklet and filtered out good tracklet based on their scores. A detection propagation method is designed to transfer the detections in the good tracklets to the bad ones so as to repair bad tracklets. The tracklet quality evaluation in our method is implemented by intra-tracklet detection consistency and inter-tracklet detection completeness. Two propagation methods; global propagation and local propagation are defined to achieve more accurate tracklet propagation. We demonstrate the effectiveness of the proposed method on the MOT 15 dataset

In this paper, we propose a novel self-supervised learning of video representation which is capable to anticipate the video category by only reading its short clip. The key idea is that we employ the Siamese convolutional network to model the self-supervised feature learning as two different image matching problems. By using frame encoding, the proposed video representation could be extracted from different temporal scales. We refine the training process via a motion-based temporal segmentation strategy. The learned representations for videos can be not only applied to action anticipation, but also to action recognition. We verify the effectiveness of the proposed approach on both action anticipation and action recognition using two datasets namely UCF101 and HMDB51. The experiments show that we can achieve comparable results with the state-of-the-art self-supervised learning methods on both tasks.

Small group detection is still a challenging problem in crowds. Traditional methods use the trajectory information to measure pairwise similarity which is sensitive to the variations of group density and interactive behaviors. In this paper, we propose two types of information by simultaneously incorporating trajectory and interaction information, to detect small groups in crowds. The trajectory information is used to describe the spatial proximity and motion information between trajectories. The interaction information is designed to capture the interactive behaviors from video sequence. To achieve this goal, two classifiers are exploited to discover interpersonal relations. The assumption is that interactive behaviors often occur in group members while there are no interactions between individuals in different groups. The pairwise similarity is enhanced by combining the two types of information. Finally, an efficient clustering approach is used to achieve small group detection. Experiments show that the significant improvement is gained by exploiting the interaction information and the proposed method outperforms the state-of-the-art methods.

Traditional action recognition approaches use pre-defined rigid areas to process the space-time information, e.g. spatial pyramids, cuboids. However, most action categories happen in an unconstrained manner, that is, the same action in different videos can happen at different places. Thus we need a better video representation to deal with the space-time variations. In this paper, we introduce the idea of mining spatial temporal saliency. To better handle the uniqueness of each video, we use a space-time over-segmentation approach, e.g. supervoxel. We choose three different saliency measures that take not only the appearance cues, but also the motion cues into consideration. Furthermore, we design a category-specific mining process to find the discriminative power in each action category. Experiments on action recognition datasets such as UCF11 and HMDB51 show that the proposed spatial temporal saliency video representation can match or surpass some of the state-of-the-art alternatives in the task of action recognition.