Dual-channel CNN for efficient abnormal behavior identification through crowd feature engineering

Abstract

His current research interests include distributed control algorithms, distributed information fusion, cooperative control, model predictive control, and machine learning. He has published more than 180 papers in international conferences and journals. His research has been supported by Royal Society, EPSRC, EU FP7, British Council, and industries. He is a board member of International Journal of Model, Identification, and Control, Cognitive Computations, Intelligent Industrial Systems, and Frontiers Robotics and AI. He served as a member of organizing committee and programme committee for many international conferences. Prof. Gu is a senior member of IEEE.

On-farm welfare monitoring system for goats based on Internet of Things and machine learning

“…Deep learning allows computational models consisting of multiple hierarchical layers to learn fantastically complex, subtle, and abstract representations instead of traditional handcrafted local descriptors and discriminative classifiers. [19][20][21][22] In the past several years, deep learning methods have been increasingly attracting research attention in many computer vision application areas. It has been shown that these techniques have provided significant improvement for object detection, including animal behavior.…”

“…In order to achieve real-time anomaly detection, Xu et al proposed an autoregressive data-driven model for environmental data streams and its corresponding prediction interval. 21 To deal with the missing data and outliers excluding, the authors introduced smooth Gaussian prior of weak assumptions on typical agricultural data, and short-term forecasts based on the Gaussian process were adopted to fill with missing data, and its prediction error was used to detect outliers. 26 Obviously, an effective and efficient anomaly detection technique not only identifies anomaly data online with high detection accuracy and low false alarm, but also satisfies some constraints in terms of computational and memory complexity.…”

One-Class Classification in Images and Videos Using a Convolutional Autoencoder With Compact Embedding

“…However, CNNs are specially designed for supervised classification problems and are not directly applicable to OCC problems. However, recent works [12], [24], [25], and [26] have shown that both, Stacked Denoising Autoencoders (SDAE) and Convolutional Autoencoders (CAE) can be alternatives for OCC problems. This is possible because they are trained to minimize the Reconstruction Error (RE) of the normal class, and this error can be used as a classification score.…”

Guest editorial: special issue on human abnormal behavioural analysis

“…The experimental evaluation showed the superiority of their proposed method over the existing stateof-the-art techniques. The same problem was studied by Xu et al in their work Dual-Channel CNN for Efficient Abnormal Behavior Identification through Crowd Feature Engineering [16]. The authors proposed a deep learningbased solution that is feed with raw video and with a motion-based spatiotemporal descriptor, resulting in a twochannel convolutional network that learns to detect abnormal behaviour in crowds.…”

“…Zhang et al introduced a method for reducing the effect of perspective distortion present in real surveillance scenarios [16]. In their article Detection of Abnormal behavior in Narrow Scene with Perspective Distortion, the authors introduced a mechanism for making up the distorting effect in the region of interest extraction followed by improved pyramid L-K optical flow method with perspective weight and disorder coefficient.…”