Deep Learning-Based Pedestrian Detection in Autonomous Vehicles: Substantial Issues and Challenges

Abstract

In recent years, autonomous vehicles have become more and more popular due to their broad influence over society, as they increase passenger safety and convenience, lower fuel consumption, reduce traffic blockage and accidents, save costs, and enhance reliability. However, autonomous vehicles suffer from some functionality errors which need to be minimized before they are completely deployed onto main roads. Pedestrian detection is one of the most considerable tasks (functionality errors) in autonomous vehicles to prevent accidents. However, accurate pedestrian detection is a very challenging task due to the following issues: (i) occlusion and deformation and (ii) low-quality and multi-spectral images. Recently, deep learning (DL) technologies have exhibited great potential for addressing the aforementioned pedestrian detection issues in autonomous vehicles. This survey paper provides an overview of pedestrian detection issues and the recent advances made in addressing them with the help of DL techniques. Informative discussions and future research works are also presented, with the aim of offering insights to the readers and motivating new research directions.

Active learning-based pedestrian and road sign detection

“…39 While traffic sign detection may be less complicated, reliable pedestrian detection is extremely difficult due to potential occlusions, deformations, and low-quality, multispectral images. 40 For occlusion detection, the authors used CNN subsections such as faster R-CNN, YOLO (You Only Look Once), and MobileNet-SSD to train the network using the complete pedestrian. The dataset was labeled into components like arm, limb, head, and individual during the training phase, which assists the dataset in distinguishing the occluded pedestrian.…”

“…47 Low-quality, multi-spectral images for pedestrian detection under low light situations and various weather conditions can be solved using different DL approaches such as YOLO. 40 This YOLO algorithm is a one-stage detector and is famous for its rapid speed. The YOLO algorithm primarily utilizes the KAIST dataset as this dataset has data in a multispectral form, and this dataset is also known as the "KAIST Multi-spectral Pedestrian Dataset".…”

Lightweight multi-target detection algorithm for unmanned aerial vehicle aerial imagery

“…Meanwhile, the single shot multibox detector (SSD) series network models improve the detection speed, but the parameter quantity is too many. 17 YOLO series network models can significantly improve the detection accuracy while slightly reducing the detection speed and have been widely used by researchers. 18 Currently, the parameter quantity and floating-point operations per second (FLOPs) of both two-stage target detection algorithms and one-stage detection algorithms are too high, so the lightweight algorithms still have difficulty achieving acceptable results under limited computing resources or mobile devices.…”

“…Although the detection accuracy of the RCNN series network is continuously improved, its detection speed is too slow to achieve real-time detection. Meanwhile, the single shot multibox detector (SSD) series network models improve the detection speed, but the parameter quantity is too many 17 . YOLO series network models can significantly improve the detection accuracy while slightly reducing the detection speed and have been widely used by researchers 18 .…”

Research on Front-Vehicle taillight recognition based on image processing and matching

“…Spatial features are captured through VGG16 [18] network, and finally time features are extracted through LSTM [19] to identify vehicle turning signals. Iftikhar S et al [20] reviewed pedestrian detection problems and the latest progress in solving these problems with the help of distance learning technology, and also introduced the informational discussion and future research work. The recall rate increased from 0.855 to 0.917.…”