ResnetCrowd: A Residual Deep Learning Architecture for Crowd Counting,\n Violent Behaviour Detection and Crowd Density Level Classification

2017/05/30 by Mark Marsden, Marsden, Mark, Kevin McGuinness +5
Computer Science · #Anomaly Detection Techniques and Applications #Video Surveillance and Tracking Methods #Human Pose and Action Recognition

paper · pdf · doi:10.48550/arxiv.1705.10698

Abstract

In this paper we propose ResnetCrowd, a deep residual architecture for\nsimultaneous crowd counting, violent behaviour detection and crowd density\nlevel classification. To train and evaluate the proposed multi-objective\ntechnique, a new 100 image dataset referred to as Multi Task Crowd is\nconstructed. This new dataset is the first computer vision dataset fully\nannotated for crowd counting, violent behaviour detection and density level\nclassification. Our experiments show that a multi-task approach boosts\nindividual task performance for all tasks and most notably for violent\nbehaviour detection which receives a 9 % boost in ROC curve AUC (Area under the\ncurve). The trained ResnetCrowd model is also evaluated on several additional\nbenchmarks highlighting the superior generalisation of crowd analysis models\ntrained for multiple objectives.\n

Related

• Crowd Scene Analysis using Deep Learning Techniques
• MTCNET: Multi-task Learning Paradigm for Crowd Count Estimation
• An Improved Dilated Convolutional Network for Herd Counting in Crowded\n Scenes
• A Deeply-Recursive Convolutional Network for Crowd Counting
• ADCrowdNet: An Attention-injective Deformable Convolutional Network for Crowd Understanding