Arindam Das

Sushil Sharma, Arindam Das, Ganesh Sistu et al.

Predicting ego vehicle trajectories remains a critical challenge, especially in urban and dense areas due to the unpredictable behaviours of other vehicles and pedestrians. Multimodal trajectory prediction enhances decision-making by considering multiple possible future trajectories based on diverse sources of environmental data. In this approach, we leverage ResNet-50 to extract image features from high-definition map data and use IMU sensor data to calculate speed, acceleration, and yaw rate. A temporal probabilistic network is employed to compute potential trajectories, selecting the most accurate and highly probable trajectory paths. This method integrates HD map data to improve the robustness and reliability of trajectory predictions for autonomous vehicles.

Sushil Sharma, Arindam Das, Ganesh Sistu et al.

Trajectory prediction is, naturally, a key task for vehicle autonomy. While the number of traffic rules is limited, the combinations and uncertainties associated with each agent's behaviour in real-world scenarios are nearly impossible to encode. Consequently, there is a growing interest in learning-based trajectory prediction. The proposed method in this paper predicts trajectories by considering perception and trajectory prediction as a unified system. In considering them as unified tasks, we show that there is the potential to improve the performance of perception. To achieve these goals, we present BEVSeg2TP - a surround-view camera bird's-eye-view-based joint vehicle segmentation and ego vehicle trajectory prediction system for autonomous vehicles. The proposed system uses a network trained on multiple camera views. The images are transformed using several deep learning techniques to perform semantic segmentation of objects, including other vehicles, in the scene. The segmentation outputs are fused across the camera views to obtain a comprehensive representation of the surrounding vehicles from the bird's-eye-view perspective. The system further predicts the future trajectory of the ego vehicle using a spatiotemporal probabilistic network (STPN) to optimize trajectory prediction. This network leverages information from encoder-decoder transformers and joint vehicle segmentation.

Saheli Hazra, Sudip Das, Rohit Choudhary et al.

Applying pseudo labeling techniques has been found to be advantageous in semi-supervised 3D object detection (SSOD) in Bird's-Eye-View (BEV) for autonomous driving, particularly where labeled data is limited. In the literature, Exponential Moving Average (EMA) has been used for adjustments of the weights of teacher network by the student network. However, the same induces catastrophic forgetting in the teacher network. In this work, we address this issue by introducing a novel concept of Reflective Teacher where the student is trained by both labeled and pseudo labeled data while its knowledge is progressively passed to the teacher through a regularizer to ensure retention of previous knowledge. Additionally, we propose Geometry Aware BEV Fusion (GA-BEVFusion) for efficient alignment of multi-modal BEV features, thus reducing the disparity between the modalities - camera and LiDAR. This helps to map the precise geometric information embedded among LiDAR points reliably with the spatial priors for extraction of semantic information from camera images. Our experiments on the nuScenes and Waymo datasets demonstrate: 1) improved performance over state-of-the-art methods in both fully supervised and semi-supervised settings; 2) Reflective Teacher achieves equivalent performance with only 25% and 22% of labeled data for nuScenes and Waymo datasets respectively, in contrast to other fully supervised methods that utilize the full labeled dataset.

Arindam Das

In the field of autonomous driving, camera sensors are extremely prone to soiling because they are located outside of the car and interact with environmental sources of soiling such as rain drops, snow, dust, sand, mud and so on. This can lead to either partial or complete vision degradation. Hence detecting such decay in vision is very important for safety and overall to preserve the functionality of the "autonomous" components in autonomous driving. The contribution of this work involves: 1) Designing a Deep Convolutional Neural Network (DCNN) based baseline network, 2) Exploiting several network remodelling techniques such as employing static and dynamic group convolution, channel reordering to compress the baseline architecture and make it suitable for low power embedded systems with nearly 1 TOPS, 3) Comparing various result metrics of all interim networks dedicated for soiling degradation detection at tile level of size 64 x 64 on input resolution 1280 x 768. The compressed network, is called SoildNet (Sand, snOw, raIn/dIrt, oiL, Dust/muD) that uses only 9.72% trainable parameters of the base network and reduces the model size by more than 7 times with no loss in accuracy

B Ravi Kiran, Arindam Das, Senthil Yogamani

Background-Foreground classification is a well-studied problem in computer vision. Due to the pixel-wise nature of modeling and processing in the algorithm, it is usually difficult to satisfy real-time constraints. There is a trade-off between the speed (because of model complexity) and accuracy. Inspired by the rejection cascade of Viola-Jones classifier, we decompose the Gaussian Mixture Model (GMM) into an adaptive cascade of Gaussians(CoG). We achieve a good improvement in speed without compromising the accuracy with respect to the baseline GMM model. We demonstrate a speed-up factor of 4-5x and 17 percent average improvement in accuracy over Wallflowers surveillance datasets. The CoG is then demonstrated to over the latent space representation of images of a convolutional variational autoencoder(VAE). We provide initial results over CDW-2014 dataset, which could speed up background subtraction for deep architectures.

Arindam Das, Olivia Das

Cloud based tiered applications are increasingly becoming popular, be it on phones or on desktops. End users of these applications range from novice to expert depending on how experienced they are in using them. With repeated usage (practice) of an application, a user's think time gradually decreases, known as learning phenomenon. In contrast to the popular notion of constant mean think time of users across all practice sessions, decrease in mean think time over practice sessions does occur due to learning. This decrease gives rise to a different system workload thereby affecting the application's short-term performance. However, such impact of learning on performance has never been accounted for. In this work we propose a model that accounts for human learning behavior in analyzing the transient (short-term) performance of a 3-tier cloud based application. Our approach is based on a closed queueing network model. We solve the model using discrete event simulation. In addition to the overall mean System Response Time (SRT), our model solution also generates the mean SRTs for various types (novice, intermediate, expert) of requests submitted by users at various levels of their expertise. We demonstrate that our model can be used to evaluate various what-if scenarios to decide the number of VMs we need for each tier-a VM configuration-that would meet the response time SLA. The results show that the lack of accountability of learning may lead to a selection of an inappropriate VM configuration. The results further show that the mean SRTs for various types of requests are better measures to consider in VM allocation process in comparison to the overall mean SRT.