InceptionDepth-wiseYOLOv2: improved implementation of YOLO framework for pedestrian detection

Pedestrian detection is one of the most challenging research areas in computer vision, as it involves classifying the image and localizing the pedestrian. Its applications, especially in automated surveillance and robotics, are exceedingly sought-after. Compared to traditional hand-crafted methods, convolutional neural networks (CNNs) have superior detection results. The single-stage detection networks, particularly the You Only Look Once (YOLO) network, have attained a satisfactory performance in object detection without compromising the computation speed and are among the state-of-the-art CNN-based methods. The YOLO framework can be leveraged to use in pedestrian detection as well. In this work, we propose an improved YOLOv2, called InceptionDepth-wiseYOLOv2. The proposed model uses a modified DarkNet53 engineered for a robust feature formation. Three inception depth-wise convolution modules are integrated at varying levels in DarkNet53, leading to a comprehensive feature of an object in the image. The proposed method is compared with state-of-the-art detection methods, i.e., FasterRCNN, YOLOv2 with various base networks, YOLOv3, and Single Shot Multibox Detector. Detection Error Trade-off Curve, Precision–Recall Curve, Log Average Miss Rate, and Average Precision performance metrics are used to compare the methods. The analysis for the count of pedestrians detected concerning their height is also carried out. The experimental study used three benchmark pedestrian datasets: the INRIA Pedestrian, PASCAL VOC 2012, and Caltech Pedestrian.