Smart cities rely on monitoring information gathered by a large number of Wireless Sensor Networks (WSN) for decision and policymaking. Recently, mobile crowd sensing (MCS), leveraging smartphones, and vehicles' latest sensing and networking features, has led to many new smart city monitoring applications. However, both WSN and MCS sensing technologies come with their challenges. WSN faces network latency, limited battery lifetime problems, security vulnerabilities, and limited capabilities due to constrained small IoT devices. On the other hand, most current MCS applications use a direct Internet connection to send the collected data to the server using 4G or 5G networks. However, this will incur an extra cost in terms of data tariff, network bandwidth, and high battery consumption. Many researches have focused on studying each of these sensing paradigms as separate architectures. However, very few studies have considered the integration between them (MCS-WSN). The integration will allow them to leverage their advantages and overcome their drawbacks. Therefore, this thesis aims to address both technologies' challenges by proposing a routing protocol implemented in both MCS and WSN nodes and allowing integration between them. The thesis also proposes a new realtime Intrusion Detection System (IDS) based on deep learning that addresses some security attacks that might happen at the WSN network. We have first investigated the routing protocols used in the IoT and smart city domain, in which the Routing Protocol for Low-Power and Lossy Networks (RPL) is selected as the base of our proposed protocol. RPL is an IETF IPv6 standard routing protocol designed for static WSN, and it is considered to be the primary routing protocol for IoT devices. We have then proposed a routing protocol named MCS-RPL that runs on MCS nodes utilizing smartphone sensing and communication capabilities. MCS-RPL uses a 2D gridbased clustering to address the RPL mobility issues and reduce control packets. Elected cell-head sensor nodes are responsible for collecting data within their cells and forwarding them in a cell-by-cell multi-hop manner to the sink node for analysis. The protocol performance was evaluated against two known MANET routing protocols: AODV and GPSR, as a baseline, and native RPL. The conducted performance evaluation reveals that the proposed protocol outperforms the others regarding packet delivery ratio and overhead packet reduction. To further improve the protocol performance, we have proposed a context-aware routing protocol that adapts to the dynamic node context in terms of status and mobility. The proposed protocol uses a new cell-head selection technique and a dynamic trickle algorithm to enhance the tree stability and further reduce the number of control packets, end-to-end delay, and increase delivery ratio shown in the evaluation. vi We have then proposed a hybrid routing protocol that combines the two wireless sensing technologies WSN and MCS, and allows the integration. The aim is to have an integrated ad-hoc hybrid-sensing network that opportunistically uses MCS nodes to support static WSN nodes to enhance the performance. The evaluation of the proposed protocol was conducted in static WSN nodes to study the impact of the integration on the WSN performance. The results reveal a good enhancement on packet delivery ratio (17 % more), end-to-end delay (50% less), and power consumption (25% less) compared to native RPL (without MCS integration). We have designed and implemented simulation models for all the routing protocols and conducted extensive simulation experiments using the OMNET++ simulation tool. We have considered using different scenarios in those experiments by varying the network density, nodes speed, and traffic load. Finally, we have proposed an IDS system based on deep-learning to address the IoT (WSN) networks' security challenge (attacks). The aim also not to transfer these threats to other nodes such as MCS and RSUs when integrating with them. Due to the lack of datasets designed for IoT traffics, we have contributed by developing a specialized IoTR dataset to detect better and classify three common attacks (DIS, Rank, and Wormhole) that target the RPL protocol. The IDS system then uses a hybrid deep learning technique that combines supervised and semi-supervised approaches to classify known and unknown anomaly behaviors within the dataset. The evaluation results show an accuracydetecting rate of 98% and 92% f1-score of multi-class attacks when using pre-trained attacks (known) and an average accuracy of 95% and 87% f1-score when predicting untrained two attack behaviors (unknown)