A Robust Framework for Traffic Object Detection using Intelligent Techniques

Object identirication is a method for pinpointing the exact position of particular items in still or moving visual media. The fields of machine learning and deep learning have resulted in a plethora of object recognition algorithms, such as SSD, SSP-net, SVM, CNN, R-CNN, Fast R-CNN, Faster R-CNN, HOG, R-FCN, YOLO, etc. While these models have been used for several object detection purposes, the difriculty of detecting tiny objects remains. Creating a lightweight and reliable object detection model that can accurately recognize little things is crucial. This research proposes an improved version of the You Only Look Once (YOLO) algorithm for identifying vehicles in surveillance footage. We replaced the YOLOv2’s original base network with DenseNet and reduced the number of parameters used. In our new and better model, we used the DenseNet-121 method to draw out the key characteristics of each picture. In addition, the compact nature of our suggested model is a result of the dense design of the underlying network. Since all of the nodes of DenseNet-121 are directly connected, we were able to use it as our foundational network and use its strength to glean insights from the lowest levels of the network and pass them on to the more advanced ones. We trained the proposed model using data obtained from Kaggle and KITTI, and we cross validated its results on the MS COCO and Pascal VOC datasets. We conducted extensive experiments to evaluate the suggested model, and the results show that our algorithm achieves higher accuracy (98.51%) than state-of-the-art methods for detecting vehicles.