keyboard_arrow_up
Accepted Papers
Novel Approach, Multi Model Video Supervision for Safety and Awareness

Lokesh Kumar1, AnubhavSingh2, 1Senior Staff Engineer @Nagarro, Dehradun, India, 2Associate Data Scientist, @Mathco, Chandigarh, India

ABSTRACT

In the modern era, surveillance systems, notably Closed-Circuit Television (CCTV) technology, have undergone a significant evolution through the integration of computer vision algorithms and deep learning methodologies, enabling them to carry out a multitude of tasks and objectives. However, a considerable portion of these systems remains far from achieving the distinction of 'smart' CCTV. In response, we introduce an innovative approach that harnesses multi-model techniques to enhance the effectiveness of CCTV supervision across a wide array of scenarios. The omnipresence of CCTV systems, functioning around the clock, leads to the accumulation of vast volumes of data, a substantial portion of which eventually becomes redundant. The conventional approach requires human supervisors to painstakingly review captured frames and subsequently deduce information pertaining to the events within those frames. This process, apart from being labour-intensive, is also time-consuming. Our novel multi-model approach represents a transformative paradigm shift within this landscape. It not only enables the system to comprehend visual data but also provides real-time analysis, alerts, and action-based responses. This advancement carries the potential to revolutionise the utilisation of CCTV systems, optimising their role in surveillance and security operations. As the world continues to witness the widespread deployment of CCTV technology, our research highlights the paradigm shift brought about by multi-model approaches, offering a pathway towards the development of intelligent, efficient, and proactive surveillance systems. This work marks our initial step in exploring the multi-modal capabilities required to understand, reason, and formulate sequences of action to enhance the current state of CCTV technology. To this end, we first conduct experiments on 87 tasks using various LLMs, including Flan-T5-Large, Vicuna, Llama 2, BLOOM, GPT-3.5, and GPT-4. These experiments show LLMs have a grasp of reasoning and can understand a scenario and act on it.

KEYWORDS

CCTV, Large Language Models, Surveillance, Alert Generation, Image to Text & Question and Answers on multiple images.


Semantic Segmentation of Road Traffic Sign Based on Improved Deeplabv3+

Wang Huifengl1 and Wu Jianfeng2, 1School of Electronic &Control Engineering Chang’an University, Xi’an 710064, Shaanxi, Chin, 2School of Information Engineering Chang’an University, Xi’an 710054, Shaanxi, China

ABSTRACT

Road traffic sign is a important facility to manage traffic and indicate the direction of traveling to ensure smooth road and driving safety. However, sign is usually small target, which is prone to the problem of missed detection and false detection. In addition, the deeplabv3+ model is a representative semantic segmentation network. It employs dilated convolution in the ASPP module, which is prone to lose small target information while expanding the sensory field. Aiming at above problem, an improved deepswin small target semantic segmentation based on deeplabv3+ is proposed. First of all, the ASPP module is replaced by swin-transformer block to enhance the signage feature extraction capability. Then, a channel and spatial fusion attention(CSFA) mechanism based on CBAM attention mechanism is utilized to enhance the extraction capability of channel and spatial features. The experimental results show that compared with the original network, the MIOU and mPA of the network proposed in this paper are improved by 4.0% and 4.9%, respectively.

KEYWORDS

Road traffic sign, deep learning, semantic segmentation, swin-transformer, attention mechanism.


Use of Dendrograms to Confirm % of Suspicions in Values That Do Not Comply With Benford's Law and Specifically Locate The Data Elements Considered Suspicious

Carlos Carrion R, Department of Data Science Technology, Andres Bello High Tech Institute, Quito, Ecuador

ABSTRACT

It is estimated that only 12% of financial/tax fraud and less than 30% of electoral fraud worldwide are detected with clear evidence, making it likely that intentional anomalies will increase aided by Artificial Intelligence and commercial access with quantum processors, since the amount of economic benefits and power is tempting, because it involves bigData, trying to clean tracks and eliminate evidence that certain systems and applications allow or do not consider taking care of. In Monitoring Audit of the normal operations behavior with help of Data Science and Technological advances, the variants of Benford's Law are available as a starting point at general level to know percentage of suspicion of values that do not comply with it; therefore, a technique is important to confirm these suspicions by more than 95% and specifically locate them through Dendrograms so that the data elements that produced suspicion are analyzed in depth, which is reason for this methodological proposal.

KEYWORDS

Types of Fraud, bigData, Artificial Intelligence, Benford Law, Dendrograms, Normal Operations Behavior.


Notorious Failure of Ms Excel With Accounting Spreadsheet in 2 Conditional Functions Giving Incorrect Results in All Versions/idioms, Reported Since 2011 That Still Persists.

Carlos Carrion R, Department of Data Science Technology, Andres Bello High Tech Institute, Quito, Ecuador

ABSTRACT

Millions of business and organizations users in the world use MS Excel spreadsheet every hour and have kept their records with calculations without considering there is any slight difference in results even if key decisions are made, such as in accounting area, when using 2 widely used conditional functions (SUMIF, COUNTIF) that in their main comparison parameter may have overflowing if it is floating point, which is used to save storage and processing due to the presence of large volume of data for the accounting action of business operations. Even more so today, with AI and social networks, millions of records are generated in business operations, it justify to abbreviate storage of business identification by country/city or IP addresses (IPv4/IPv6) where it can easily exceed 15 digits, to arrange them in floating point and use conditional functions in spreadsheet to group or differentiate them, which with this failure is sure to make wrong decisions.

KEYWORDS

Accounting Spreadsheet, Wrong Results, Artificial Intelligence, Business Operations, Software Failure.


menu
Reach Us

emailcseai@cseai2024.org


emailcseaiconf@yahoo.com

close