A method for obstacle avoidance based on TOF sensing and binocular vision
By equipping unmanned devices with TOF sensing modules and binocular cameras, and combining light intensity and distance conditions, the weights are dynamically adjusted to perform obstacle distance fusion, solving the problems of ranging errors and false alarms in low-light and low-texture scenes for unmanned devices, and achieving accurate obstacle avoidance.
Patent Information
- Application Number
- CN202610331750.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-18
- Publication Date
- 2026-06-23
AI Technical Summary
Existing forward-looking binocular vision systems for unmanned devices suffer from decreased ranging accuracy in low-light environments, misjudging obstacles and failing to meet the requirements for precise obstacle avoidance. They perform poorly, especially in close-range and low-texture scenes. Furthermore, the TOF perception and binocular vision fusion solution has failed to effectively solve the false alarm problem.
By equipping unmanned devices with a TOF sensing module, binocular camera, and photosensitive module, and combining TOF distance, binocular distance, and ambient light intensity, the weights W1 and W2 are dynamically adjusted to perform obstacle distance fusion calculation, thereby optimizing ranging accuracy and obstacle avoidance strategies.
Under different lighting and distance conditions, it significantly improved ranging accuracy and obstacle avoidance accuracy, reduced false alarm and false alarm rates, and met the obstacle avoidance requirements of unmanned equipment.
Smart Images
Figure CN122260342A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine vision technology, and in particular to an obstacle avoidance method based on TOF perception and binocular vision. Background Technology
[0002] Currently, unmanned equipment (including drones, unmanned vehicles, and unmanned boats) widely adopts forward-looking binocular vision systems for obstacle avoidance decisions, but it has the following drawbacks:
[0003] 1. When the ambient light intensity is <100 lux, the signal-to-noise ratio of the binocular camera's photosensitive element drops to below 20 dB, which leads to a significant increase in the parallax matching failure rate and ranging error, and may even result in the complete inability to output effective distance data.
[0004] 2. Due to the limitations of the optical characteristics of binocular cameras, when faced with low-texture scenes such as "glass reflection, metal mirror, and solid color wall", the binocular vision system is prone to misjudging them as obstacles, while missing low-contrast targets such as "small branches and wires".
[0005] 3. At close range, such as within 5 meters, due to the limitation of baseline length (usually ≤150mm), the parallax calculation accuracy of the binocular vision system has a large deviation, which cannot meet the requirements for accurate obstacle avoidance and obstacle circling.
[0006] Meanwhile, existing technologies have solutions that fuse TOF (Time of Flight) perception with binocular vision data, but these solutions only focus on fusing the distance measurements of the two, without considering the impact of ambient light intensity on the binocular vision system, and without establishing a verification mechanism for false alarms from the binocular data system by TOF perception. As a result, the false alarm problem cannot be fundamentally solved, and the decision-making requirements for accurate obstacle avoidance cannot be met. Summary of the Invention
[0007] The purpose of this invention is to provide an obstacle avoidance method based on TOF perception and binocular vision. This method can assign values to the TOF distance weight W1 and the binocular distance weight W2 based on different obstacle distance ranges and ambient lighting conditions to perform different fusion calculations, so as to give full play to the ranging performance of the TOF perception module and the binocular camera at different distances and meet different task execution requirements.
[0008] To achieve the above objectives, the present invention provides the following technical solution:
[0009] An obstacle avoidance method based on Time-of-Flight (TOF) perception and binocular vision is provided, which includes the following steps:
[0010] Equip unmanned devices with a TOF sensing module, a binocular camera, and a photosensitive module;
[0011] During the movement of the unmanned device, the TOF sensing module and the binocular camera synchronously and correspondingly acquire the distance of the same obstacle, which are respectively denoted as TOF distance D_TOF and binocular distance D_Stereo. At the same time as the TOF sensing module and the binocular camera synchronously acquire the distance of the obstacle, the photosensitive module acquires the light intensity L of the current environment in real time.
[0012] Based on the TOF distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L, the corresponding obstacle distance determination strategy is executed to obtain the final obstacle distance D_final;
[0013] Furthermore, the unmanned equipment executes an obstacle avoidance strategy based on the final obstacle distance D_final.
[0014] Preferably, the obstacle distance determination strategy is executed based on the TOF distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L to obtain the final obstacle distance D_final, including:
[0015] If the current ambient light intensity L < the light intensity threshold, then the TOF distance D_TOF is directly used as the final obstacle distance D_final;
[0016] If the current ambient light intensity L ≥ the light intensity threshold, then dynamic distance fusion is performed according to D_final = W1 × D_TOF + W2 × D_Stereo to obtain the final obstacle distance D_final, where W1 and W2 are the TOF distance weight and the binocular distance weight, respectively, and W1 + W2 = 1.
[0017] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0018] If the TOF distance D_TOF < M1 and the binocular distance D_Stereo < M1, then set W1=1 and W2=0 to calculate the final obstacle distance D_final.
[0019] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0020] If M1 ≤ TOF distance D_TOF < M2, and M1 ≤ binocular distance D_Stereo < M2, then take W1 = [0.8, 0.9] and W2 = [0.1, 0.2] to calculate the final obstacle distance D_final.
[0021] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0022] If M2 ≤ TOF distance D_TOF < M3, and M2 ≤ binocular distance D_Stereo < M3, then take W1=[0.4, 0.6] and W2=[0.4, 0.6] to calculate the final obstacle distance D_final.
[0023] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0024] If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| ≤ distance difference threshold Δd, and the confidence level S_TOF of the TOF distance D_TOF ≥ confidence threshold S, then take W1=[0.4, 0.6] and W2=[0.4, 0.6] to calculate the final obstacle distance D_final.
[0025] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0026] If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| > distance difference threshold Δd, and the confidence level S_TOF of TOF distance D_TOF ≥ confidence threshold S, then take W1=1 and W2=0 to calculate the final obstacle distance D_final.
[0027] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0028] If M3 ≤ TOF distance D_TOF and M3 ≤ binocular distance D_Stereo, then take W1=0 and W2=1 to calculate the final obstacle distance D_final.
[0029] Preferably, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0030] If M2≤TOF distance D_TOF<M3, and based on the binocular images obtained by the binocular camera, it is considered that there are no obstacles, and the confidence level S_TOF of the TOF distance D_TOF≥the confidence threshold S, then W1=1 and W2=0 are taken to calculate the final obstacle distance D_final.
[0031] If M3 ≤ TOF distance D_TOF, and based on the binocular images obtained by the binocular camera, it is assumed that there are no obstacles, and the confidence level S_TOF of the TOF distance D_TOF ≥ the confidence threshold S, then W1=1 and W2=0 are used to calculate the final obstacle distance D_final.
[0032] Preferably, the unmanned device executes an obstacle avoidance strategy based on the final obstacle distance D_final, including:
[0033] If the final obstacle distance D_final is less than the first distance threshold H1, the unmanned device will stop moving and output an alarm message. The value of H1 is in the range of [2.5m, 3.5m].
[0034] If the first distance threshold H1 ≤ final obstacle distance D_final < second distance threshold H2, the unmanned device decelerates and executes an obstacle avoidance procedure. The value range of H2 is [20m, 25m].
[0035] If the third distance threshold H2 ≤ final obstacle distance D_final < third distance threshold H3, the unmanned device will decelerate and change its direction of movement. The value range of H2 is [40m, 50m].
[0036] In summary, the present invention has the following advantages compared with the prior art:
[0037] This invention can assign values to the TOF distance weight W1 and the binocular distance weight W2 based on different obstacle distance ranges and ambient lighting conditions, and further perform different fusion calculations to fully utilize the ranging performance of the TOF sensing module and the binocular camera at different distances to meet different task execution requirements. At the same time, it comprehensively considers the obstacle results output by the binocular camera to significantly reduce false alarms and missed alarms caused by relying on the ranging results of the binocular camera. Attached Figure Description
[0038] Figure 1 This is a flowchart illustrating the steps of the obstacle avoidance method based on TOF perception and binocular vision in this invention. Detailed Implementation
[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0040] Example 1:
[0041] like Figure 1 As shown, this embodiment provides an obstacle avoidance method based on TOF perception and binocular vision, which includes the following steps:
[0042] S1. Equip the unmanned device with a TOF sensing module 1, a binocular camera 2, and a photosensitive module 3; In this embodiment, the unmanned device includes any one of a drone, an unmanned vehicle, and an unmanned boat.
[0043] S2. During the movement of the unmanned equipment, the TOF sensing module 1 and the binocular camera 2 synchronously and correspondingly acquire the distance of the same obstacle, which are respectively denoted as TOF distance D_TOF and binocular distance D_Stereo, and the time stamp synchronization accuracy of the TOF sensing module 1 and the binocular camera 2 when acquiring the distance of the obstacle is ≤1ms.
[0044] In addition, while the TOF sensing module 1 and the binocular camera 2 simultaneously acquire the distance to the obstacle, the photosensitive module 3 acquires the light intensity L of the current environment in real time;
[0045] S3. Execute the corresponding obstacle distance determination strategy based on the TOF distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L to obtain the final obstacle distance D_final;
[0046] Additionally, S4, the unmanned equipment executes an obstacle avoidance strategy based on the final obstacle distance D_final.
[0047] Specifically, in step S3, the corresponding obstacle distance determination strategy is executed based on the TOF distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L, including:
[0048] If the current ambient light intensity L < the light intensity threshold (e.g., 100 lux), then the TOF distance D_TOF is directly used as the final obstacle distance D_final. This is because when the current ambient light intensity L < the light intensity threshold, the stereo camera 2 is in a low-light environment, and its frame rate drops significantly, such as to 10 fps. At this point, the stereo camera 2 is almost ineffective, so there is no need to consider the influence of the stereo distance D_Stereo on the final obstacle distance D_final. The light intensity threshold can be set according to the technical parameters of the stereo camera 2. For example, under normal circumstances, the light adaptation range of the stereo camera 2 is 100-10000 lux, so the light intensity threshold can be set to 100 lux.
[0049] If the current ambient light intensity L ≥ the light intensity threshold (e.g., 100 lux), then dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, where W1 and W2 are the TOF distance weight and the binocular distance weight, respectively, and W1+W2=1;
[0050] Furthermore, dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0051] If the TOF distance D_TOF < M1 and the binocular distance D_Stereo < M1, then W1=1 and W2=0 are used to calculate the final obstacle distance D_final. That is, the TOF distance D_TOF is directly used as the final obstacle distance D_final. The value range of M1 is [1m, 2m] (unit: meters). Since the distance measurement accuracy of the TOF sensing module 1 can reach the millimeter level within the distance range less than the first distance threshold M1, under this close-range condition, the difference in the viewing angle between the left and right cameras of the binocular camera 2 is too large, and feature matching is difficult, which leads to a decrease in the ranging accuracy of the binocular camera 2 or even measurement failure. Therefore, under this condition, the TOF distance D_TOF is directly used as the final obstacle distance D_final.
[0052] If M1 ≤ TOF distance D_TOF < M2, and M1 ≤ binocular distance D_Stereo < M2, then W1 = [0.8, 0.9] and W2 = [0.1, 0.2] are used to calculate the final obstacle distance D_final, and the value range of M2 is [4m, 6m] (unit: meters). Since both the TOF distance D_TOF and the binocular distance D_Stereo satisfy the above conditions of M1 and M2, and under good lighting conditions, the parallax and baseline of the binocular camera 2 reach an ideal balance, which can play the best measurement performance. Therefore, high ranging accuracy (accuracy can reach the centimeter level) can be obtained. Therefore, under this medium distance condition, the ranging accuracy of the binocular camera 2 needs to be taken into account, but the ranging result position of the TOF sensing module 1 is still used (because at this time it is still close-range ranging, and the ranging accuracy advantage of the TOF sensing module 1 is obvious). Therefore, a large weight W1 (for example, W1 = 0.9) and a small weight W2 (for example, W2 = 0.1) are used for fusion calculation.
[0053] If M2 ≤ TOF distance D_TOF < M3, and M2 ≤ binocular distance D_Stereo < M3, then W1 = [0.4, 0.6] and W2 = [0.4, 0.6] are used to calculate the final obstacle distance D_final, and the value range of M3 is [20m, 25m] (unit: meters). When both the TOF distance D_TOF and the binocular distance D_Stereo satisfy the above conditions M2 and M3, and under sufficient lighting conditions, the binocular camera 2 can obtain high-quality scene images, and based on the scene images of this quality, high ranging accuracy (accuracy up to centimeter level) can be obtained. Therefore, under this distance condition, the ranging accuracy of the binocular camera 2 needs to be taken into account, but the ranging result position of the TOF sensing module 1 is still used (because at this time it is still close-range ranging, and the ranging accuracy advantage of the TOF sensing module 1 is obvious). Therefore, weights W1 and W2 (for example, W2 = 0.1) are used for fusion calculation.
[0054] If M3 ≤ TOF distance D_TOF and M3 ≤ binocular distance D_Stereo, then W1=0 and W2=1 are used to calculate the final obstacle distance D_final. Within this long-distance range and under sufficient lighting conditions, the binocular distance D_Stereo is directly used as the final obstacle distance D_final to fully utilize the advantages of the binocular camera 2 in terms of its excellent imaging quality and ability to achieve large field of view target tracking and positioning, thereby completing the target recognition and tracking of high-speed moving targets (such as aircraft).
[0055] Therefore, in this embodiment, under sufficient lighting conditions (i.e., the current ambient light intensity L ≥ the light intensity threshold (e.g., 100 lux)), the TOF distance weight W1 and the binocular distance weight W2 can be set based on different distance ranges, and different fusion calculations can be performed to give full play to the ranging performance of the TOF sensing module 1 and the binocular camera 2 at different distances, so as to meet different task execution requirements.
[0056] For example, within a distance range of 1-5 meters, the distance measurement accuracy using only the binocular camera 2 is ±0.3 meters, while after calculation using the fusion algorithm in this embodiment, the distance measurement accuracy is ±0.3 meters, which is a 67% improvement compared to using only the binocular camera 2.
[0057] For example, within this long-distance range (i.e., when both the TOF distance D_TOF and the binocular distance D_Stereo are ≥ M3), and under sufficient lighting conditions, the binocular distance D_Stereo can be directly used as the final obstacle distance D_final to achieve large field-of-view target tracking and positioning (at this time, the ranging accuracy of the binocular camera 2 does not need to be considered too much, but only its advantages of good long-distance imaging effect and accurate target recognition can be utilized).
[0058] Example 2:
[0059] The only difference between this embodiment and Embodiment 1 is that dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0060] If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| ≤ distance difference threshold Δd (Δd ranges from [0.3m, 0.5m]), and the confidence level S_TOF of the TOF distance D_TOF ≥ confidence threshold S (S ranges from [0.6, 0.9]), then we take W1 = [0.4, 0.6] and W2 = [0.4, 0.6] to calculate the final obstacle distance D_final, and the range of M3 is [20m, 25m] (unit: meters).
[0061] If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| > distance difference threshold Δd, and the confidence level S_TOF of TOF distance D_TOF ≥ confidence threshold S, then it is considered that the ranging result of binocular camera 2 has a false alarm. In this case, W1=1 and W2=0 are taken to calculate the final obstacle distance D_final, that is, the TOF distance D_TOF is directly used as the final obstacle distance D_final.
[0062] Example 3:
[0063] The only difference between this embodiment and Embodiment 1 is that dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including:
[0064] If M2≤TOF distance D_TOF<M3, and based on the binocular images acquired by binocular camera 2, it is assumed that no obstacle exists (D_Stereo can be considered to be infinite at this time), and the confidence S_TOF of the TOF distance D_TOF≥confidence threshold S (the value of S is in the range of [0.6, 0.9]), then it is considered that the ranging result of binocular camera 2 has a false alarm. At this time, W1=1 and W2=0 are used to calculate the final obstacle distance D_final, that is, the TOF distance D_TOF is directly used as the final obstacle distance D_final.
[0065] If M3 ≤ TOF distance D_TOF, and based on the binocular images acquired by binocular camera 2, it is considered that no obstacle exists, and the confidence level S_TOF of the TOF distance D_TOF ≥ the confidence threshold S (the value of S is in the range of [0.6, 0.9]), then it is considered that the ranging result of binocular camera 2 has a missed detection. At this time, W1=1 and W2=0 are taken to calculate the final obstacle distance D_final, that is, the TOF distance D_TOF is directly used as the final obstacle distance D_final.
[0066] Therefore, embodiments 2-3 above can take into account the confidence level of D_TOF and the ranging results of the binocular camera 2, i.e. whether there is an obstacle, in order to significantly reduce the occurrence of false alarms and missed alarms caused by relying on the ranging results of the binocular camera 2. For example, the false alarm rate can be reduced from 15% to less than 3%, and the missed alarm rate can be reduced from 8% to less than 1%.
[0067] Example 4:
[0068] The only difference between this embodiment and any of embodiments 1-3 is that, in step S4, the unmanned device executes an obstacle avoidance strategy based on the final obstacle distance D_final, including:
[0069] If the final obstacle distance D_final < the first distance threshold H1, the unmanned device stops moving (such as when a drone hovers) and outputs an alarm message. The value of H1 is in the range of [2.5m, 3.5m] (unit: meters).
[0070] If the first distance threshold H1 ≤ final obstacle distance D_final < second distance threshold H2, the unmanned device decelerates and executes an obstacle avoidance procedure. The value range of H2 is [20m, 25m] (unit: meters).
[0071] If the third distance threshold H2 ≤ final obstacle distance D_final < third distance threshold H3, the unmanned device will decelerate and change its direction of movement. The value range of H2 is [40m, 50m] (unit: meters).
[0072] In summary, this invention can assign values to the TOF distance weight W1 and the binocular distance weight W2 based on different obstacle distance ranges and ambient lighting conditions, and further perform different fusion calculations to fully leverage the ranging performance of the TOF sensing module and the binocular camera at different distances to meet different task execution requirements. At the same time, it comprehensively considers the obstacle results output by the binocular camera, i.e., whether or not obstacles exist, to significantly reduce false alarms and missed alarms caused by relying on the ranging results of the binocular camera.
[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An obstacle avoidance method based on TOF perception and binocular vision, characterized in that, Includes the following steps: Equip unmanned devices with a TOF sensing module, a binocular camera, and a photosensitive module; During the movement of the unmanned device, the TOF sensing module and the binocular camera synchronously and correspondingly acquire the distance of the same obstacle, which are respectively denoted as TOF distance D_TOF and binocular distance D_Stereo. At the same time as the TOF sensing module and the binocular camera synchronously acquire the distance of the obstacle, the photosensitive module acquires the light intensity L of the current environment in real time. Based on the TOF distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L, the corresponding obstacle distance determination strategy is executed to obtain the final obstacle distance D_final; Furthermore, the unmanned equipment executes an obstacle avoidance strategy based on the final obstacle distance D_final.
2. The obstacle avoidance method as described in claim 1, characterized in that, Based on the Time-of-Flight (TOF) distance D_TOF, the binocular distance D_Stereo, and the current ambient light intensity L, the corresponding obstacle distance determination strategy is executed to obtain the final obstacle distance D_final, including: If the current ambient light intensity L < the light intensity threshold, then the TOF distance D_TOF is directly used as the final obstacle distance D_final; If the current ambient light intensity L ≥ the light intensity threshold, then dynamic distance fusion is performed according to D_final = W1 × D_TOF + W2 × D_Stereo to obtain the final obstacle distance D_final, where W1 and W2 are the TOF distance weight and the binocular distance weight, respectively, and W1 + W2 = 1.
3. The obstacle avoidance method as described in claim 2, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If the TOF distance D_TOF < M1 and the binocular distance D_Stereo < M1, then set W1=1 and W2=0 to calculate the final obstacle distance D_final.
4. The obstacle avoidance method as described in claim 3, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M1 ≤ TOF distance D_TOF < M2, and M1 ≤ binocular distance D_Stereo < M2, then take W1 = [0.8, 0.9] and W2 = [0.1, 0.2] to calculate the final obstacle distance D_final.
5. The obstacle avoidance method as described in claim 3, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M2 ≤ TOF distance D_TOF < M3, and M2 ≤ binocular distance D_Stereo < M3, then take W1=[0.4, 0.6] and W2=[0.4, 0.6] to calculate the final obstacle distance D_final.
6. The obstacle avoidance method as described in claim 3, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| ≤ distance difference threshold Δd, and the confidence level S_TOF of the TOF distance D_TOF ≥ confidence threshold S, then take W1=[0.4, 0.6] and W2=[0.4, 0.6] to calculate the final obstacle distance D_final.
7. The obstacle avoidance method as described in claim 3, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M2 ≤ TOF distance D_TOF < M3, M2 ≤ binocular distance D_Stereo < M3, and |D_TOF - D_Stereo| > distance difference threshold Δd, and the confidence level S_TOF of TOF distance D_TOF ≥ confidence threshold S, then take W1=1 and W2=0 to calculate the final obstacle distance D_final.
8. The obstacle avoidance method according to any one of claims 5-7, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M3 ≤ TOF distance D_TOF and M3 ≤ binocular distance D_Stereo, then take W1=0 and W2=1 to calculate the final obstacle distance D_final.
9. The obstacle avoidance method according to any one of claims 5-7, characterized in that, Dynamic distance fusion is performed according to D_final=W1×D_TOF+W2×D_Stereo to obtain the final obstacle distance D_final, including: If M2≤TOF distance D_TOF<M3, and based on the binocular images obtained by the binocular camera, it is considered that there are no obstacles, and the confidence level S_TOF of the TOF distance D_TOF≥the confidence threshold S, then W1=1 and W2=0 are taken to calculate the final obstacle distance D_final. If M3 ≤ TOF distance D_TOF, and based on the binocular images obtained by the binocular camera, it is assumed that there are no obstacles, and the confidence level S_TOF of the TOF distance D_TOF ≥ the confidence threshold S, then W1=1 and W2=0 are used to calculate the final obstacle distance D_final.
10. The obstacle avoidance method as described in claim 1, characterized in that, The unmanned device executes an obstacle avoidance strategy based on the final obstacle distance D_final, including: If the final obstacle distance D_final is less than the first distance threshold H1, the unmanned device will stop moving and output an alarm message. The value of H1 is in the range of [2.5m, 3.5m]. If the first distance threshold H1 ≤ final obstacle distance D_final < second distance threshold H2, the unmanned device decelerates and executes an obstacle avoidance procedure. The value range of H2 is [20m, 25m]. If the third distance threshold H2 ≤ final obstacle distance D_final < third distance threshold H3, the unmanned device will decelerate and change its direction of movement. The value range of H2 is [40m, 50m].