A Superficial Learning Approach to Detect Criminal Activities in Roadsides using Hybrid Deep Learning Mechanism

However, the Crime activity video/image materials from these cameras cannot be accurately assessed only by visual observation and manual video analysis. In order to overcome limitations of conventional systems, we need a solution able to classify captured video and photos as well as provide active assistance to surveillance workers. Therefore, this study proposes an improved deep learning methodology (SLA-HDL), which makes a crime detection system. The feature extraction part of this scheme is comprised of several layers advanced methodology that is required to extract the features and classified from the video frames. Apart from the proposed system for criminal activity detection, two deep learning methodologies such as Modified Yolo3 and DeepNet-V3 are trained and evaluated on both real time and collected crime scene photo and video. The proposed detection system includes input choices for video preparation and video enhancement, dataset of crime activity pictures, application of the proposed deep learning method as well as a few other pre-trained methods. Our suggested model must be accurate with 99.54%, precision is 99.37%, a recall of 99.45%, and false positive rate is 0.342 as shown in the studies. The data yielded by our proposed model are superior to other traditional models.