The development of the Internet of Vehicles (IoV) offers significant opportunities to enhance the traffic management system based on connected vehicles (CVs), while accurate traffic arrival rate estimation is critical for the dynamic evaluation and optimization of signalized intersections. Existing CV-based methods, however, are constrained to single-stream estimation that assumes first-in-first-out (FIFO) discipline, overlook initial queues, and deteriorate sharply when CV penetration is low or data are spoofed. To address these limitations, this study proposes a joint maximum a posteriori (JO-MAP) method that jointly estimates cycle-based arrival rates of multiple traffic streams under both undersaturated and oversaturated conditions. The key innovations include a joint weighted likelihood function that treats each queued CV as an independent observation, eliminates FIFO assumptions, and explicitly accounts for the initial queues, and a joint Bayesian prior that embeds historical CV sample-size information for enhanced accuracy even with sparse real-time CV data. Comprehensive simulation and field experiments show that JO-MAP produces reliable estimates under different penetration rates, arrival patterns, and traffic volume levels, achieving 100% estimation success and less-than 4 veh cycle-level error with only 5% CV penetration. The feature of joint estimation makes the method less demanding for the penetration rate of CVs and more robust to noisy/spoofing data compared to baseline methods, limiting the error increase to 1.2 veh under deliberate spoofing attacks. Besides, JO-MAP reduces average vehicle delay by 12%-20% when integrated into adaptive signal control, demonstrating its potential for IoV-enabled traffic management.