Abstract-Urban target recognition at intersection using multimodality wireless sensor networks is very promising in reducing accidents by detecting unusual events in real time. To provide alarm signal about the incoming car to the pedestrian using sound, or notify the car driver about the pedestrian using infrastructure-to-vehicle communication, the deployed sensor system collects the sense data from multimodality sensor nodes, performs data fusion, and conducts reactions to avoid imminent accident. To address the problem, we design and implement App-MAC to support prioritized event delivery, provide interevent and intra-event fairness, improve the performance of channel utilization, and reduce energy consumption. App-MAC leverages the advantages of contention-based and reservation-based MAC protocols to coordinate the channel access, and propose channel contention and reservation algorithms to adaptively allocate the time slots according to application requirements and current events status. To evaluate App-MAC, we have conducted simulations through TOSSIM simulator and empirical studies using Berkeley TelosB motes with target recognition events, and compared with three state-of-the-art MAC protocols, i.e., S-MAC, TDMA, and TRAMA, in terms of the proposed performance metrics, namely average event delivery latency, event and sensor fairness index, channel utilization efficiency, and energy consumption efficiency. We found that App-MAC outperforms other protocols tremendously in this application scenario.