Depthwise Convolution for Compact Object Detector in Nighttime Images

Heena Patel, Kalpesh Prajapati, Anjali Sarvaiya, Kishor Upla, Kiran Raja, Raghavendra Ramachandra, Christoph Busch; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 379-389


Despite thermal imaging primarily used for night-time surveillance, uniform temperature of object and background makes it difficult to acquire details in the scene being observed and thereby object detection. Further, motion objects exhibit noisy features, and thermal images collected over long distances degrade the spatial resolution of the acquired objects. We present a computationally efficient object detection approach using Depthwise Deep Convolutional Neural Network (DDCNN) for detecting and classifying objects for night-time images under low resolution. The Depthwise Convolution (DC) employed in the proposed object detection algorithm minimises the network's computational complexity resulting in lowest number of training parameters (i.e., 3M) as compared to the other existing stateof-the-art methods such as FRCNN (52M), SSD (24M) and YOLO-v3 (61M) parameters. Further, through introducing novel Tversky and Intersection over Union (IoU) loss functions into the compact architectural design, we improves night-time object detection accuracy. The validity of the proposed model is assessed on numerous datasets such as FLIR, KAIST, MS, and our internal datasets having multiple objects. The experimental results using proposed method indicate both quantitative and qualitative improvements over the recent state-of-the-art methods for nighttime imaging. The proposed approach achieves a mean Average Precision (mAP) of 52.39% and a highest individual object detection accuracy of 72.70% accuracy for cars in nigh-time situations suggesting applications in real-time use cases.

Related Material

[pdf] [supp]
@InProceedings{Patel_2022_CVPR, author = {Patel, Heena and Prajapati, Kalpesh and Sarvaiya, Anjali and Upla, Kishor and Raja, Kiran and Ramachandra, Raghavendra and Busch, Christoph}, title = {Depthwise Convolution for Compact Object Detector in Nighttime Images}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2022}, pages = {379-389} }