Today, wrongdoings are expanding step by step and will be intense. There are numerous weapon detection algorithms such as Hybrid weapon detection algorithm, using material test and fuzzy logic system, automated handgun detection, YOLOV3 object detection algorithm which can detect the weapons There are some traditional classifiers, for example, Naive Bayes, SVM, K-Nearest, and so forth in AI. Be that as it may, these calculations have a few constraints. The proposed work will conquer constraint of these calculations by building a custom classifier which will give more precision than in traditional classifiers. This work will be useful to maintain a strategic distance from the bank thefts, Crime at public spots, at ATM's, and so on. The primary goal of this exploration paper is to construct ongoing covered weapon and small object detection in video and building custom classifier that orders the action into suspicious and non-suspicious acitivity. The significant advantage of this work is for criminal justice in AI.