A Comparative Study of Pedestrian Detection Techniques Over the Last Decade

Pedestrian detection has made significant strides in the recent decade because of breakthroughs in the fields of deep learning. Computer vision-based applications, from object detection and tracking to surveillance footage and, most frequently, driverless cars, rely on pedestrian identification. Human detection in streets and pathways is an important aspect of various highly important jobs. There are numerous approaches to this problem, including the use of digital cameras, infrared or heat-detecting devices, time-of-flight sensors, and so on. All of them function by gathering data about the environment and then predicting or estimating where a person is concerning the sensor. In the assessment, the efficiency of detection and tracking is evaluated using several image sources such as RGB, thermal, and multispectral formats. The proposed research delves into R-CNN, Fast R-CNN, Single Shot detector, and Scale-Aware Fast R-CNN approaches. These methods assess the works by utilizing datasets from KITTI, Caltech, and INRIA. These results are compared to over 20 other earlier research conducted utilizing the same datasets. With the KITTI, Caltech, and INRIA datasets, Faster R-CNN gets the greatest average precision of all pedestrian identification models tested.