As cities continue to densify, intelligent connected vehicles increasingly encounter practical challenges in parking—especially in gated residential communities where tight layouts and diverse obstacles demand higher-performing autonomous parking solutions. To address autonomous parking in narrow residential parking bays, this study presents a path-planning approach built on multi-sensor fusion localization. An environmental sensing platform is developed using 12 ultrasonic sensors and four high-definition cameras, and a fusion framework is constructed by combining a camera model, an IMU measurement model, and a wheel-speed (tachometer) kinematic model. An enhanced inverse-expansion Hybrid A* planning method is introduced to boost efficiency by swapping the start and goal positions, allowing node expansion to proceed from the constrained interior space toward a more open area. Simulation results indicate that planning completes within 1.4 seconds across scenarios, with a best-case runtime of 0.75 seconds. Parking-space feasibility tests show that at 3 km/h the minimum required space is 6.821 m × 2.164 m, increasing to 7.058 m × 2.205 m at 6 km/h. The method achieves safe planning for both perpendicular and parallel parking, while keeping the vehicle’s intersection-position error relative to the parking boundary within 12 cm. Overall, the proposed approach offers a practical and effective technical pathway for autonomous parking in complex residential settings.