Faster GSAC-DNN: A Deep Learning Approach to Nighttime Vehicle Detection Using a Fast Grid of Spatial Aware Classifiers




Faster GSAC-DNN (Grid of Spatial Aware Classifiers based on Deep Neural Networks) is an evolution of the detection algorithms GSAC and GSAC-DNN with faster and more efficient performance. Faster GSAC-DNN updates the concept of Grid of Spatial-Aware Classifiers with a single convolution operation capable of extracting local feature maps adapted to the spatial position of each classifier. The predictions made by the grid of classifiers are later post-prcessed to infer a set of point-based detections. For applications such as nighttime vehicle detection, where light flashes may occlude the real shape of vehicles, point-based detections are more appropriate than the classical bounding boxes since it can be difficult to fit the size of the boxes to the observed light flashes. For that reason, Faster GSAC-DNN has been tested on a nighttime
traffic dataset like
NVD (Nighttime Vehicle Detection database), which already provides point- based annotations.


For questions about this neural network, please contact Daniel Fuertes at This email address is being protected from spambots. You need JavaScript enabled to view it.


Click here to download the code.

Click here to download the Pretrained weights on the NVD database.



Daniel Fuertes, Carlos R. del-Blanco, Fernando Jaureguizar, Narciso García, “A Deep Learning Approach to Nighttime Vehicle Detection Using a Fast Grid of Spatial Aware Classifiers”, submitted to International Conference on Image Processing (ICIP), 2023.