Crowd Density Analysis and Suspicious Activity Detection
Shriya Akella, Priyanka A. Abhang, Vinit Agrharkar, Reena Sonkusare · 2020 IEEE International Conference for Innovation in Technology (INOCON) · 2020
There is a growing need for smart surveillance systems in public and private places to logically differentiate between normal and abnormal behaviour. This is not just important for the convenience of the people but also for their security. Understanding a video footage and classifying an activity as normal or suspicious especially in densely packed regions is possible and has been demonstrated in this paper. The proposed system makes use of the YOLOv3 algorithm for object detection. The COCO dataset, which is a large-scale object detection, segmentation and captioning dataset, has been used for training this model. The whole framework is made up of two parts. First the features are computed from the image. Then based on the detected features, the classifier makes a prediction. Depending on the object detected, the algorithm classifies a frame as suspicious or normal. Crowd density has been calculated by detecting the number of people in a frame and suspicion detection has been performed by analysing a frame for suspicious objects like isolated bags, knives and guns.