Suspicious Activity Detection using LRCN

The massive rise in criminal activity in urban and suburban regions and suspicious behavior in public places pose a considerable risk to people’s safety and social order. It is vital to minimize those activities. Previously, humans used to monitor these activities manually, which is a very strenuous task. So, this article focuses on constructing a model for behavior recognition understanding and classifying the incredulous activity using artificial intelligence and deep learning concepts based on the LRCN model by combining CNNs and LSTMs. These concepts are helpful in learning visual features from video frames and to transfer a sequence of image embedding into class labels, probabilities, and sentences. As a result, this model has the benefit of both convolutional and recurrent neural networks. The resulting model detects suspicious activities that fall under 13 different categories, including bullying, firing, attack, robbery, blast, violence, extortion, gunshot, filching, annexation, and rifling. The proposed innovation categories the events using the LRCN model in python with Keras and tensor flow. This model collects frames from the videos and compares them to trained models with accessibility and high functionality with an accuracy of about 91.54929% making the model practically possible in solving an ample number of cases with limited resources.