In this paper, we present a novel approach to video understanding through the introduction of an Omnidirectional Linear Attention Module (OLAM). This module effectively captures and processes spatial, temporal, and channel information in video data using linear transformer concepts. Traditional attention mechanisms often struggle with the computational complexity of processing high-dimensional video data, particularly in capturing long-range dependencies. Our OLAM addresses these challenges by leveraging linear transformers, which reduce computational complexity from quadratic to linear with respect to sequence length. To support this research, we introduce the Football Penalty Shootout Video Dataset (FPSD), comprising 2426 high-quality videos from top-tier tournaments. These videos feature subtle player movements, including feints and deceptive tactics, presenting ultimate challenges for video understanding. The complexity of the background in these videos introduces extraneous information, further complicating the analysis. To effectively analyze this challenging dataset, we employ triplet loss in pretraining designed specifically for handling complex video datasets along three separate dimensions. This module is integrated into our reimagined lightweight Omnidirectional Linear Attention Network (OLANet), which excels in deciphering temporal information from short videos, demonstrating a remarkable capability to focus on key details amidst the high noise of FPSD. Our approach, combining OLANet with triplet loss pretraining, significantly reduces computational complexity without compromising performance, achieving state-of-the-art accuracy of approximately 96% on the FPSD.