This paper presents an improved event-based stereo visual odometry (ESVO) system that addresses the limitations of the original ESVO framework in terms of accuracy and efficiency. The main contributions include: 1. **Adaptive Accumulation (AA) of Events**: A novel method for efficient edge pixel sampling based on the local dynamics of events, reducing computational complexity and improving mapping completeness and smoothness. 2. **Temporal and Static Stereo Mapping**: Combining temporal and static stereo results to enhance depth recovery and mapping accuracy. 3. **IMU-Aided Camera Pose Tracking**: Introducing gyroscope measurements as a prior to improve the accuracy of yaw component recovery in general 6-DoF motion. The proposed system is evaluated on two publicly available datasets (rpg and DSEC), demonstrating superior performance in mapping completeness, depth smoothness, and trajectory accuracy compared to the original ESVO. The system also shows better computational efficiency, making it suitable for real-time applications with high-resolution event cameras.This paper presents an improved event-based stereo visual odometry (ESVO) system that addresses the limitations of the original ESVO framework in terms of accuracy and efficiency. The main contributions include: 1. **Adaptive Accumulation (AA) of Events**: A novel method for efficient edge pixel sampling based on the local dynamics of events, reducing computational complexity and improving mapping completeness and smoothness. 2. **Temporal and Static Stereo Mapping**: Combining temporal and static stereo results to enhance depth recovery and mapping accuracy. 3. **IMU-Aided Camera Pose Tracking**: Introducing gyroscope measurements as a prior to improve the accuracy of yaw component recovery in general 6-DoF motion. The proposed system is evaluated on two publicly available datasets (rpg and DSEC), demonstrating superior performance in mapping completeness, depth smoothness, and trajectory accuracy compared to the original ESVO. The system also shows better computational efficiency, making it suitable for real-time applications with high-resolution event cameras.