Self-Driving Cars Perception Pedestrian Detection

Autonomous vehicles (AVs) have been increasingly adopted for their potential to improve overall road safety and reduce traffic congestion. However, many significant challenges remain to be overcome before AVs can safely navigate busy urban environments, such as accurately perceiving their surroundings to avoid potential collisions. One of the most major tasks that provide the most value, and have a profound impact on AVs perception system progress is detecting pedestrians and vehicles, both in day and night conditions. This chapter reviews the recent advances in computer vision research, also known as vision-based techniques which use visual feedback information extracted from a vision sensor to detect and recognize pedestrians and vehicles, and even to control the motion of a vehicle, in an AV perception system. Traditional and deep learning (DL) methods are discussed in detail, highlighting their advantages and limitations in both day and night conditions. The review concludes that DL techniques provide the best results for the detection of pedestrians and vehicles, although small, occluded, and truncated objects are still hard to detect. Further research is recommended on new and challenging datasets, to attain the utmost performance of the algorithms in highly uncertain environments understanding, like in dense surroundings and difficult light and weather conditions.