The change of crowd energy is a fundamental measurement for describing a crowd behavior. In this paper, we present a crowd abnormal detection method based on the change of energy-level distribution. The method can not only reduce the camera perspective effect, but also detect crowd abnormal behavior in time. Pixels in the image are treated as particles, and the optical flow method is adopted to extract the velocities of particles. The qualities of different particles are distributed as different value according to the distance between the particle and the camera to reduce the camera perspective effect. Then a crowd motion segmentation method based on flow field texture representation is utilized to extract the motion foreground, and a linear interpolation calculation is applied to pedestrian’s foreground area to determine their distance to the camera. This contributes to the calculation of the particle qualities in different locations. Finally, the crowd behavior is analyzed according to the change of the consistency, entropy and contrast of the three descriptors for co-occurrence matrix. By calculating a threshold, the timestamp when the crowd abnormal happens is determined. In this paper, multiple sets of videos from three different scenes in UMN dataset are employed in the experiment. The results show that the proposed method is effective in characterizing anomalies in videos.

The general paradigm of traditional Siamese networks involves using cross-correlations to fuse features from the backbone, and this paradigm is limited by the inductive bias of the convolution kernel, resulting in the lack of global information. In this paper, we propose the Siamese learning regression distribution (SiamLRD) to address the local limitations of traditional cross-correlation operations on feature fusion and weak self-connections between features within different branches. The SiamLRD uses the cross-attention mechanism to replace cross-correlations between the features of the target region of interest and the template so as to enhance flexibility. Firstly, the original transformer structure is improved to be suitable for convolutional Siamese networks. The improved transformer architecture is then used to replace cross-correlation operations, resulting in more comprehensive feature fusion between branches. Secondly, we introduce a new decoder structure into the novel fusion strategy to enhance the correlation between classification scores and regression accuracy during decoding. Multiple benchmarks are used to test the proposed SiamLRD approach, and it is verified that the proposed approach improves the baseline with 5.8% in terms of AO and 9.7% in terms of SR0.75 on the GOT-10K dataset.

Anchor-free based trackers introduce an extra branch in addition to classification and regression branches in the network to achieve comparable performance with anchor-based trackers. This extra branch is usually trained independently in the training phase and is used in combination with other branches in the inference phase. However, this can increase the inconsistency between the inference phase and the training phase, potentially degrading the tracking performance. To address this problem, we propose a new Siamese network-based object tracking framework that eliminates this inconsistency by unifying classification and additional branch tasks to achieve learning location quality estimation. Furthermore, regression tasks for bounding boxes are widely formulated based on Dirac $\delta $ distribution. Though this assumption works well for many scenarios, it restricts the prediction of regression branches. To overcome this restriction, we propose discretizing the continuous offset of the regression branch into multiple offset predictions, which enables the network to learn more flexible distributions automatically. Meanwhile, the discrete distribution prediction of regression branches is utilized to further guide the classification of the trackers. Extensive experiments on the widely accepted benchmarks demonstrate the effectiveness and efficiency of the proposed model.

Aiming at the problem of poor robustness and the low effectiveness of target tracking in complex scenes by using single color features, an object-tracking algorithm based on dual color feature fusion via dimension reduction is proposed, according to the Correlation Filter (CF)-based tracking framework. First, Color Name (CN) feature and Color Histogram (CH) feature extraction are respectively performed on the input image, and then the template and the candidate region are correlated by the CF-based methods, and the CH response and CN response of the target region are obtained, respectively. A self-adaptive feature fusion strategy is proposed to linearly fuse the CH response and the CN response to obtain a dual color feature response with global color distribution information and main color information. Finally, the position of the target is estimated, based on the fused response map, with the maximum of the fused response map corresponding to the estimated target position. The proposed method is based on fusion in the framework of the Staple algorithm, and dimension reduction by Principal Component Analysis (PCA) on the scale; the complexity of the algorithm is reduced, and the tracking performance is further improved. Experimental results on quantitative and qualitative evaluations on challenging benchmark sequences show that the proposed algorithm has better tracking accuracy and robustness than other state-of-the-art tracking algorithms in complex scenarios.