Cooperative perception allows a Connected Autonomous Vehicle (CAV) to interact with the other CAVs in the vicinity to enhance perception of surrounding objects to increase safety and reliability. It can compensate for the limitations of the conventional vehicular perception such as blind spots, low resolution, and weather effects. An effective feature fusion model for the intermediate fusion methods of cooperative perception can improve feature selection and information aggregation to further enhance the perception accuracy. We propose adaptive feature fusion models with trainable feature selection modules. One of our proposed models Spatial-wise Adaptive feature Fusion (S-AdaFusion) outperforms all other State-of-the-Arts (SO-TAs) on two subsets of the OPV2V dataset: Default CARLA Towns for vehicle detection and the Culver City for domain adaptation. In addition, previous studies have only tested cooperative perception for vehicle detection. A pedestrian, however, is much more likely to be seriously injured in a traffic accident. We evaluate the performance of cooperative perception for both vehicle and pedestrian detection using the CODD dataset. Our architecture achieves higher Average Precision (AP) than other existing models for both vehicle and pedestrian detection on the CODD dataset. The experiments demonstrate that cooperative perception also improves the pedestrian detection accuracy compared to the conventional single vehicle perception process.

Methods of estimating heart rate without the use of sensor devices provides essential benefits in both the medical field as well as the other computing applications. Smartphones are the handiest devices available to everyone today. By using videos of fingertip captured with smartphone camera, heart rate (HR) can be estimated using the photoplethysmography (PPG) technique. It is based on tracking subtle color changes on the skin owing to cardiovascular activities. These color changes are invisible to the human eye but can be detected by digital cameras. The method is divided into three main steps: first, reading the video frames and processing them to obtain the PPG data, next, extracting the Blood Volume Pulse (BVP) signal, and finally, estimating the HR from the signal. In this project, the color intensity of the skin pixels is used, and filters are applied to eliminate the noise and retain only the pulses of interest. The extracted signal is fed into a convolutional regression neural network which outputs the estimated HR. The results obtained are compared with the ground truth HR obtained by using a contact PPG sensor. We obtained a Mean Absolute Error (MAE) of 7.01 beats per minute (bpm) and an error percentage of 8.3% on test data.

Drivable area or free space detection is an important task in Advanced Driver-Assistance Systems (ADAS) and autonomous driving system. It can help intelligent vehicles understand road conditions and determine safe driving area. Semantic segmentation is a pixel-wise prediction which can classify each pixel into its category. In this paper, we propose a deep learning-based semantic segmentation architecture to predict the drivable area in front of the vehicle. Our model is built based on ResNet backbone with the Feature Pyramid Network (FPN) and Atrous Spatial Pyramid Pooling (ASPP) modules. The backbone in the bottom-up architecture extracts features and an ASPP is attached to the last decoder layer. Additionally, a top-down architecture with lateral connections is added in the decoder and the FPN utilizes the multi-scale features for final prediction. Our model is evaluated on the Cityscapes street scene dataset and achieves 95.90% mIoU on road segmentation. Next, the model is evaluated on the BDD100K large-scale diverse driving dataset with direct drivable region and alternative drivable region annotations. For this dataset our model achieves 84.58% mIoU which is comparable to some State-of-the-Art models.

Vital signs are important parameters that can reflect people's physiological status and help physicians provide medical advice. Remote Photoplethysmography (rPPG) is a fast, low-cost and convenient method to remotely collect biometric data, and requires only a facial video recorded using a smartphone or other camera. Remote medical service provisioning proved to be a dire need during the COVID-19 pandemic. To leverage the cloud-based medical advice provisioning platform of Your Doctors Online, we propose a rPPG methodology to measure people's Heart Rate (HR) and Heart Rate Variability (HRV) based on a facial video recorded by the users using a smartphone. We validate our model on the TokyoTech remote PPG dataset.

During all trimesters of the pregnancy, measuring the fetal Head Circumference (HC) from ultrasound images can estimate the gestational age, monitor the growth status of the fetus and infer newborn's state. Precise segmentation of fetal ultrasound images can help physicians measure HC efficiently and accurately and make further predictions. In this paper, we leverage deep learning encode-decode architecture to segment the fetal skull boundary and fetal skull for fetal HC measurement. We modify our network based on U-Net due to its outstanding performance in biomedical image analysis. We add dilated convolution layers after the last encoder and Squeeze-and-Excitation (SE) blocks on the skip connections of U-Net to segment fetal skull boundary and fetal skull in 2D ultrasound images. The model is trained and evaluated on the HC18 grand challenge dataset, which has 2D ultrasound images at different trimesters of pregnancy. We achieved 2.27 ± 3.61 mm mean absolute difference in HC measurement. The model also achieved 97.31 ± 1.84% mean Dice score in fetal skull segmentation.