Machine-Learning Object Detection and Recognition for Surveillance System using YoloV3

Intelligent Video surveillance is one of the most emerging technologies in Computer vision, used for object detection and locating within the video or image. Majority of the research are carried on the Yolo algorithm on vehicle tracking, monitoring vehicles, and medical science. The main objective of this paper is to develop an optimal solution to detect, locate multiple objects such as person and vehicles in a single frame using Kitti dataset. Usually the kitti data set focuses on foreground vehicle detection; in the proposed algorithm it detects person, vehicle and also the background objects. The output obtained for every image includes the information of object such as probability, classification of the object, bonding box, object center (x, y coordinates), height, width using Non-Maximum Suppression (NMS) algorithm. The Kitti dataset of 350 images is used and it is observed that classification rate is 80% at 0.3 confidence threshold value over bounding boxes pixel area for vehicle and person detection. This work can be further carried out in detecting and tracking of objects at different weather conditions like rainy, winter and also during night.