What question did this study set out to answer?

The aim is to improve pedestrian detection accuracy in complex urban environments using a machine learning framework.

February 21, 2026Open Access

Surveillance video–based pedestrian detection for smart transportation using machine learning techniques

Key Points

The aim is to improve pedestrian detection accuracy in complex urban environments using a machine learning framework.
Developed a location-aware pedestrian identification framework for smart transportation.
Utilized a cascaded convolutional architecture with semantic scene parsing.
Implemented region proposal refining for accurate pedestrian positioning.
Employed a location-perception layer combined with smart occupancy-based filtering.
Achieved higher detection accuracy compared to standard deep neural networks and region-based CNNs.
Significantly reduced false alarms in complex urban surveillance footage.
Demonstrated effectiveness in real-world applications with well-known pedestrian datasets.

Abstract

Pedestrian detection is an essential prerequisite for intelligent transportation systems and autonomous surveillance, necessitating consistent performance in challenging conditions, including occlusion, scale variation, and subpar data quality (Roszyk in Sensors, 22(3), 1082), (Gawande in Procedia Comput Sci. 218, 2438-2447, 2023), (Li in IEEEAccess 10:19937 - 19957, 2022). Detectors that use deep learning have been very successful, but they don’t always work well when there is a lot of traffic because they can’t always tell where they are or how to see the whole scene (Hajari in Proceedings of ICDIC 2020 (pp. 29-38). Singapore:Springer Nature Singapore, 2020), (Marnissi in Pattern Recognition Letters, 153:222-231, 2022). A location-aware pedestrian identification framework for surveillance-based smart transportation improves detection accuracy and contextual comprehension in urban traffic. Cascaded convolutional architecture with semantic scene parsing and region proposal refining detects pedestrians and their positions in traffic contexts like roads and walkways (Saeidi in The Visual Comput. 38(6):2223-2237, 2022), (Dasgupta in IEEE Trans Intell Transp Syst. 23(9):15940-15950, 2022). A lightweight segmentation backbone makes semantic features at different levels of detail. Then, these features are put together with candidate bounding boxes using a location-perception layer that uses a smart occupancy-based filtering method to get rid of bad proposals. The framework is tested on well-known pedestrian datasets, such as the Caltech Pedestrian and CUHK Square datasets (Islam in IEEE Trans Intell Transp Syst. 23(12):23358-23368, 2022), (Song in Neurocomputing, 482:98-115, 2022). The experimental results show that the proposed method always does better than standard deep neural network and region-based CNN baselines on the datasets that were tested. This shows that it works well for accurately recognizing pedestrians in real-world surveillance settings. Enhances pedestrian detection through scene-aware semantic filtering. Reduces false alarms in complex urban surveillance footage. Allows for smart transportation monitoring in real time.

Ask AI

Helpful

Bookmark

View Full Paper