Dynamic Occupancy Grid algorithms are an essential part of modern perception systems for autonomous vehicles, allowing for a combined representation of static and dynamic environments. However, these algorithms are known to be computationally expensive and difficult to optimize. This paper presents a detailed analysis of a radar-based DOGM algorithm, focusing on computation time and identifying key computational bottlenecks, particularly in the particle filter’s resampling step. Two variations of the resampling process are proposed, effectively reducing the mean and variance of computation times without compromising performance. These modifications result in a 7% decrease in average computation time and approximately 50% less variance for higher particle counts.