Ashmitha Sivakumar

Ms. Ashmitha Sivakumar

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 September 2025 in Machine Vision and Applications
Verified
Journal Article
Open access
PDF file
AAAayush Agrawal
Aayush Agrawal
ASAshmitha Sivakumar
Ashmitha Sivakumar
IKIbrahim Kaif
Ibrahim Kaif
CBChayan Banerjee
Chayan Banerjee
Accurate lane segmentation is essential for effective path planning and lane following in autonomous driving, especially in scenarios with significant occlusion from vehicles and pedestrians. Existing models often struggle under such conditions, leading to unreliable navigation and safety risks. We propose two innovative approaches to enhance lane detection in these challenging environments, each showing notable improvements over conventional methods. The first approach aug-Segment improves conventional lane detection models by augmenting the training dataset (e.g. CULanes) with simulated occlusions and training a segmentation model. This method achieves a 12% improvement over a number of SOTA models on the CULanes dataset. Additionally, we present a second approach, LOID: Lane Occlusion Inpainting and Detection, designed to address the issue more comprehensively and with improved robustness. LOID introduces an advanced lane segmentation network that uses an image processing pipeline to identify and mask occluded regions. An inpainting model is then applied to reconstruct the road environment in the occluded areas. The enhanced image is then processed by a lane detection algorithm, resulting in a 20% and 24% improvement over several SOTA models on the BDDK100 and CULanes datasets respectively, highlighting the effectiveness of the proposed approach.

30 August 2024 in Smart Cities
Verified
Journal Article
Open access
PDF file
ECEnrique D. Saldivar Carranza
Enrique D. Saldivar Carranza
Rahul Suryakant SakhareRahul Suryakant Sakhare
Rahul Suryakant Sakhare
JDJairaj Chetas Desai
Jairaj Chetas Desai
JMJijo K. Mathew
Jijo K. Mathew
ASAshmitha Sivakumar
Ashmitha Sivakumar
DBDarcy Bullock
Darcy Bullock
Commercially available crowdsourced connected vehicle (CV) trajectory data have recently been used to provide stakeholders with actionable and scalable roadway mobility infrastructure performance measures. Transportation agencies and automotive original equipment manufacturers (OEMs) share a common vision of ensuring the privacy of motorists that anonymously provide their journey information. As this market has evolved, the fleet mix has changed, and some OEMs have introduced additional fuzzification of CV data around 0.5 miles of frequently visited locations. This study compared the estimated Indiana market penetration rates (MPRs) between historic non-fuzzified CV datasets from 2020 to 2023 and a 5–11 May 2024, CV dataset with fuzzified records and a reduced fleet. At selected permanent interstate and non-interstate count stations, overall CV MPRs decreased by 0.5% and 0.3% compared to 2023, respectively. However, the trend in previous years was upward. Additionally, this paper evaluated the impact on data characteristics at freeways and intersections between the 5–11 May 2024, fuzzified CV dataset and a non-fuzzified 7–13 May 2023, CV dataset. The analysis found that the total number of GPS samples decreased 10% statewide. Of the evaluated 54,284 0.1-mile Indiana freeway, US Route, and State Route segments, the number of CV samples increased for 33.8% and decreased for 65.9%. This study also evaluated 26,291 movements at 3289 intersections and found that the number of available trajectories increased for 28.3% and decreased for 70.4%. This paper concludes that data representativeness is enough to derive most relevant mobility performance measures. However, since the change in available trajectories is not uniformly distributed among intersection movements, an unintended sample bias may be introduced when computing performance measures. This may affect signal retiming or capital investment opportunity identification algorithms.

17 August 2024 in arXiv
Verified
Preprint
Open access
PDF file
AAAayush Agrawal
Aayush Agrawal
ASAshmitha Sivakumar
Ashmitha Sivakumar
IKIbrahim Kaif
Ibrahim Kaif
CBChayan Banerjee
Chayan Banerjee
Accurate lane detection is essential for effective path planning and lane following in autonomous driving, especially in scenarios with significant occlusion from vehicles and pedestrians. Existing models often struggle under such conditions, leading to unreliable navigation and safety risks. We propose two innovative approaches to enhance lane detection in these challenging environments, each showing notable improvements over current methods. The first approach aug-Segment improves conventional lane detection models by augmenting the training dataset of CULanes with simulated occlusions and training a segmentation model. This method achieves a 12% improvement over a number of SOTA models on the CULanes dataset, demonstrating that enriched training data can better handle occlusions, however, since this model lacked robustness to certain settings, our main contribution is the second approach, LOID Lane Occlusion Inpainting and Detection. LOID introduces an advanced lane detection network that uses an image processing pipeline to identify and mask occlusions. It then employs inpainting models to reconstruct the road environment in the occluded areas. The enhanced image is processed by a lane detection algorithm, resulting in a 20% & 24% improvement over several SOTA models on the BDDK100 and CULanes datasets respectively, highlighting the effectiveness of this novel technique.